LIGHT SPEED FUTURE

# API Integration Guide

Native Seedance Video · LSF Seedance 2.5 Production<br>
LSF Seedance 2.5 Flash · Asset Library · Seed Audio<br>
LSF Video Enhancement

Version 2.10.5  |  17 November 2026

New to LSF? Start with the First Video Quickstart to generate a 5-second, 720p, 16:9 Native Seedance 2.5 video and get its video link. Use this complete guide when you need more controls. Native Seedance 2.5 Flash is covered separately in section 5A; LSF Seedance 2.5 Flash Fast and Standard are covered in section 7. Use the exact model alias enabled for your account.

SCOPE AND ACCESS

This guide describes the customer APIs for video generation, speech generation, asset libraries and video enhancement. It covers authentication, request parameters, responses, retry behaviour and billing units. Use the model aliases and product permissions supplied by LSF for your account.

Audience: API integration teams and developers

Company website: https://lightspeedfuture.com

API base URL: https://ai-api.lightspeedfuture.com

Examples contain placeholders. Replace the API key, model alias, task or asset identifiers, and example.com media URLs with your authorized values.

## About this guide

| Item | Value |
| --- | --- |
| Revision | v2.10.5 · 17 November 2026 |
| Account access | Use the model aliases assigned to your account. Each product chapter lists its output options and request parameters. |
| Pricing | Use the effective price agreed for your account, product, tier, resolution and input mode. Public list prices and account-specific terms can differ. Duration examples explain billing units and are not quotations or spending caps. |
| Task interface | POST /v1/videos and GET /v1/videos/{task_id}; original-file GET /v1/videos/{task_id}/content for Production video outputs (section 6.7), Video Enhancement video outputs (section 10.4), and Native Flash (section 5A.4). Video callbacks and task cancel/delete/list are outside this guide. |

### Contents

- 1. Choose the Right Workflow
- 2. Connection and Authentication
- 3. Customer Onboarding Values
- 4. First Request and Safe Retry
- 5. Native Video Generation
- 5A. Native Seedance 2.5 Flash
- 6. LSF Seedance 2.5 Production
- 7. LSF Seedance 2.5 Flash
- 8. Asset Library
- 9. Seed Audio Speech Generation
- 10. LSF Video Enhancement
- 11. Integration and Error Handling
- 12. Frequently Asked Questions
- 13. Further Reading
- 14. Support
- Appendix A. Customer Integration Checklist

## 1. Choose the Right Workflow

Use the exact model alias issued for the product you intend to call. The video products share task endpoints, but their inputs, outputs and billing models differ.

| Workflow | You provide | You receive |
| --- | --- | --- |
| Native Seedance 2.0 / 2.5 | Prompt and supported image, video, audio or frame references | AI video generation with Seedance’s native creative controls. Seedance 2.5 supports Native 4K, and an optional 480p Draft preview followed by a separate 1080p or 4K Final. See section 5. |
| Native Seedance 2.5 Flash | Prompt and supported image, audio or frame references | Native 480p/720p MP4 or MOV, with completion-token billing. Use an enabled Native Flash alias; see section 5A. |
| LSF Seedance 2.5 Production | Prompt and supported generation references; select the assigned tier alias | Create videos from prompts and references, with output up to 8K. See section 6. |
| LSF Seedance 2.5<br>Flash | Prompt and supported generation references; select a Fast or Standard alias | Fast video generation with one task and a finished MP4. Default output is 1080p. See section 7. |
| Asset Library | Authorized asset registration or verification requests | Reusable asset identifiers for supported generation references. See section 8. |
| Seed Audio | Text and optional supported audio or image references | A speech-audio download URL. See section 9. |
| LSF Video Enhancement | An existing video URL and its duration | Improve existing footage, with resolution and format options for playback and editing. See section 10. |

Video workflow scope

Choose Native Seedance 2.5 for its native generation controls and optional Draft preview-to-Final workflow. Choose LSF Seedance 2.5 Production for finished video with flexible delivery controls, or LSF Seedance 2.5 Flash for fast video generation. Production and Flash each deliver the finished result through one public task; their output resolutions are distinct from the Native model’s output options.

Use the product-specific request and response fields below. Named output resolutions generally describe resolution bands; actual dimensions also depend on the source and aspect ratio. Each product chapter explains its supported dimensions, aspect ratios and frame rates.

| Product | Output selection | Request ID | Final media field |
| --- | --- | --- | --- |
| Native Seedance 2.5 | 480p / 720p / 1080p / 4k | Recommended | metadata.url |
| Native Seedance 2.5 Flash | 480p / 720p | Recommended | metadata.url |
| LSF Seedance 2.5 Production | Lite: 1080p / 2k / 4k; Standard and Pro: 1080p / 2k / 4k / 6k / 8k | Recommended | metadata.url |
| LSF Seedance 2.5 Flash | Fast up to 4k; Standard includes 6k and 8k; defaults to 1080p | Required | metadata.url |
| Seed Audio | Speech generation | Recommended | url |
| LSF Video Enhancement | Named resolution or short-side limit; required | Recommended | metadata.video_url |

## 2. Connection and Authentication

| Setting | Value |
| --- | --- |
| API base URL | https://ai-api.lightspeedfuture.com unless LSF supplies an account-specific address |
| Authentication | Authorization: Bearer &lt;LSF_API_KEY&gt; |
| Request type | Content-Type: application/json |
| Video create / retrieve | POST /v1/videos<br>GET /v1/videos/{task_id} |
| Balance | GET /v1/billing/balance |
| Speech | POST /v1/audio/speech |
| Assets | Documented /v1/seedance/virtual/... and /v1/seedance/real-human/... paths |
| Model identifiers | Model aliases supplied by LSF. Asset Library requests do not include model. |

Use the LSF request fields in this guide even if your HTTP client supports an OpenAI-compatible style. Use the parameters documented here; other vendors’ fields and SDK defaults may not be supported.

Keep API keys on your application server. Use the creating key while it remains valid. For Production and Flash, a valid replacement key issued for the same LSF organization and account can retrieve an existing task, subject to its current permissions. Retain the original task ID and recover with GET; do not resubmit the old POST under a new key. Request IDs remain scoped to each key, so switching keys can create a separate billable task. Native Draft-to-Final key rotation is described in section 5.8; do not assume these rules apply to every endpoint.

### Balance response

```text
GET /v1/billing/balance
Authorization: Bearer <LSF_API_KEY>
{
  "object": "billing.balance",
  "currency": "USD",
  "balance": {
    "available": 100.0,
    "used": 0.0,
    "unlimited": false
  },
  "updated_at": 1782493727
}
```

This endpoint reports the available and used allowance associated with the current API key, not a sum of all account keys or an account wallet. Pending reservations affect the displayed allowance; they are not necessarily final charges. For an unlimited key, available is null and unlimited is true. This does not mean the account has unlimited funds. updated_at is the response-generation time in Unix seconds. Pricing and top-ups follow your agreement with LSF.

For accepted Production and Flash tasks, an exhausted key allowance does not by itself prevent retrieval. Disabled, revoked, expired or deleted keys cannot be used; use a valid, authorized replacement key. Existing tasks retain their original billing terms, and reading with a replacement key does not transfer or repeat their charges. New submissions still require sufficient allowance and current product permissions.

## 3. Customer Onboarding Values

LSF supplies these values during onboarding. The model alias selects the product; your enabled service determines which output combinations can be submitted.

| Item | Value to obtain |
| --- | --- |
| API access | Customer / account name, API base URL and API key delivered separately |
| Native video aliases | Seedance 2.5 and any contracted Seedance 2.0 Standard / Fast / Mini aliases |
| Native Seedance 2.5 Flash alias | &lt;YOUR_NATIVE_FLASH_MODEL_ALIAS&gt;; only after LSF confirms enablement |
| LSF Seedance 2.5 Production aliases | &lt;YOUR_PRODUCTION_LITE_MODEL_ALIAS&gt;<br>&lt;YOUR_PRODUCTION_STANDARD_MODEL_ALIAS&gt;<br>&lt;YOUR_PRODUCTION_PRO_MODEL_ALIAS&gt; |
| LSF Seedance 2.5 Flash aliases | &lt;YOUR_FLASH_FAST_MODEL_ALIAS&gt;<br>&lt;YOUR_FLASH_STANDARD_MODEL_ALIAS&gt; |
| Asset access | Virtual, real-human, or both; asset identifiers authorized for your use |
| Speech | &lt;YOUR_SEED_AUDIO_MODEL_ALIAS&gt;, if contracted |
| LSF Video Enhancement | &lt;YOUR_VIDEO_ENHANCEMENT_MODEL_ALIAS&gt; for Standard and Professional; &lt;YOUR_VIDEO_ENHANCEMENT_LITE_MODEL_ALIAS&gt; for Lite |
| Enabled outputs | Product, tier and resolution combinations enabled for your account; supported input forms and last-frame availability |
| Commercial terms | Agreed prices, balance / API-key allowance, invoices and retention requirements |
| Support | Telegram @HenryBroG · info@lightspeedfuture.com |

Use the exact alias supplied by LSF; do not construct it from an account display name. GET /v1/models can show the models currently visible to your key. Model visibility does not mean every parameter combination is enabled. Use your assigned model alias and the parameters documented for that product.

## 4. First Request and Safe Retry

For the simplest first test, use the First Video Quickstart at https://lightspeedfuture.com/docs/quickstart/. It shows how to create a 5-second, 720p, 16:9 Native Seedance 2.5 video with curl and retrieve the video link. The direct curl example below uses the same settings; product-specific options follow in later chapters.

```text
curl --request POST \
  'https://ai-api.lightspeedfuture.com/v1/videos' \
  --header 'Authorization: Bearer <LSF_API_KEY>' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "<YOUR_SEEDANCE_2_5_MODEL_ALIAS>",
    "prompt": "A ceramic cup on a desk in soft daylight.",
    "metadata": {
      "duration": 5,
      "resolution": "720p",
      "ratio": "16:9",
      "output_format": "mp4",
      "generate_audio": false,
      "client_request_id": "demo-native-001"
    }
  }'
```

Store the returned public task identifier with the exact submitted request and client_request_id. Retrieve that task using the same authorized key:

```text
curl 'https://ai-api.lightspeedfuture.com/v1/videos/<YOUR_TASK_ID>' \
  --header 'Authorization: Bearer <LSF_API_KEY>'
```

| Situation | Action |
| --- | --- |
| Accepted / still processing | Poll the same task every 30 seconds or longer. LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash advance automatically even when you do not poll. |
| Completed | Read the documented result URL and copy the media to your own storage promptly. |
| Failed; or expired for Native video | Read the error and review the balance adjustment. LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash currently expose completed and failed as their terminal states. A corrected, intentionally new job uses a new request ID. |
| GET timeout or connection loss | Resume GET on the same task. Do not create another job. |
| POST outcome unknown | Keep the original request ID and payload. If you have a task ID, use GET. Do not create a fresh request ID merely because the response was lost; contact LSF if the original task cannot be established. |
| Identical idempotent replay | Reuse the original client_request_id only with the same logical request and unchanged payload. A changed URL or output setting is not an identical replay. |

The initial charge estimate is not a guarantee of final duration or a spending cap. Keep sufficient available balance for the final charge adjustment after completion.

### 4.1 Request IDs and identical replay

Assign a unique metadata.client_request_id before the first POST. Save the exact request body, the key identity used for submission and the returned public task ID. Use a fresh ID for each intentionally new job.

| Product | Requirement | Allowed ID |
| --- | --- | --- |
| Native video | Optional; recommended | 1–128 characters |
| LSF Seedance 2.5 Production | Optional; recommended | 1–128 characters |
| LSF Seedance 2.5 Flash | Required | 1–128 characters |
| Seed Audio | Optional; recommended | 1–128 characters |
| LSF Video Enhancement | Optional; recommended | 1–64 characters |

Use only letters A–Z and a–z, digits 0–9, and . _ : - . Submit IDs without leading or trailing whitespace. Video IDs share one namespace per API key across Native, LSF Seedance 2.5 Production, LSF Seedance 2.5 Flash and LSF Video Enhancement; do not reuse an ID for another product on the same key.

For LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash, a replay of an accepted identical request returns HTTP 200 with the task’s current state, rather than repeating the original HTTP 202 response. It does not create a new generation or reservation. Reusing an ID with a different request returns HTTP 409. Adding an explicit default, changing prompt whitespace or reordering reference items can change the request; replay the saved body unchanged.

Switching API keys creates a different idempotency scope and can create another billable task. Keep the original key, ID and body when a submission outcome is unknown. Once a task ID is available, resume GET instead of resubmitting to check progress.

Video deduplication depends on the accepted task record remaining available. This guide does not specify a fixed or permanent video-idempotency retention period. Native Draft eligibility, result-URL expiry and Seed Audio replay windows are separate rules.

Replay the saved request body unchanged. Do not rely on field normalization or infer one product’s replay rules from another. Seed Audio replay windows are described in section 9.8.

### 4.2 Production and Flash task responses

LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash use the response format in this section. Use the model alias supplied for the selected product. Native video, Seed Audio and LSF Video Enhancement have their own response fields, documented in their respective chapters.

| Status | Meaning | Client action |
| --- | --- | --- |
| queued | Accepted and awaiting processing | Save the task ID and poll. |
| in_progress | The workflow is running | Continue GET on the same task. |
| completed | The final result is ready | Read metadata.url and store the media. |
| failed | The workflow ended in failure | Read error and the reservation adjustment. |

The first accepted POST returns HTTP 202. An accepted identical replay returns HTTP 200 with the current task state. GET returns HTTP 200 for a readable task, including a failed one; inspect status as well as the HTTP status. Progress values do not promise a completion time.

#### Accepted task

```json
{
  "id": "task_example",
  "task_id": "task_example",
  "object": "video",
  "model": "<YOUR_PRODUCT_MODEL_ALIAS>",
  "status": "queued",
  "progress": 0,
  "created_at": 1792100000
}
```

#### Processing task

```json
{
  "id": "task_example",
  "task_id": "task_example",
  "object": "video",
  "model": "<YOUR_PRODUCT_MODEL_ALIAS>",
  "status": "in_progress",
  "progress": 10,
  "created_at": 1792100000
}
```

#### Completed task

The following result is illustrative. The actual output length, frame rate and completion time depend on the request and delivered media.

```json
{
  "id": "task_example",
  "task_id": "task_example",
  "object": "video",
  "model": "<YOUR_PRODUCT_MODEL_ALIAS>",
  "status": "completed",
  "progress": 100,
  "created_at": 1792100000,
  "completed_at": 1792100180,
  "seconds": "5.025",
  "metadata": {
    "url": "https://example.com/final.mp4",
    "duration": "5.025",
    "fps": "120",
    "resolution": "1080p"
  }
}
```

id and task_id identify the same public task. Read the final video from metadata.url. seconds, metadata.duration, metadata.fps and metadata.resolution are strings. created_at and completed_at use Unix seconds; completed_at is optional until recorded.

metadata.last_frame_url is present only when requested and successfully produced. Fields without a value are generally omitted. The response does not echo client_request_id: retain your own request-ID to task-ID mapping. Use the public task ID and the documented result fields for tracking and delivery.

#### Failed task

```json
{
  "id": "task_example",
  "task_id": "task_example",
  "object": "video",
  "model": "<YOUR_PRODUCT_MODEL_ALIAS>",
  "status": "failed",
  "progress": 100,
  "created_at": 1792100000,
  "completed_at": 1792100180,
  "error": {
    "code": "PRODUCTION_PROCESSING_FAILED",
    "message": "PRODUCTION_PROCESSING_FAILED"
  }
}
```

Both products currently return PRODUCTION_PROCESSING_FAILED for a workflow failure. Submission errors are listed in section 11.1. A confirmed workflow failure releases the reservation; an unknown submission outcome or client timeout alone does not establish failure.

Download successful media promptly and retain it in your own storage. Result links have no uniform guaranteed retention period. Production video outputs support the authenticated original-file route in section 6.7; use the documented result URL for Flash. Do not infer download support across products.

## 5. Native Video Generation

Native Seedance video generation supports text, image, video and audio references, with creative controls for duration, aspect ratio and frame composition. The API follows an OpenAI-compatible style; place Seedance-specific controls under metadata as documented below.

This chapter covers the existing Native models. The separate Native Seedance 2.5 Flash model is specified in section 5A; do not apply this chapter's flagship 4K, Draft-to-Final, reference limits or bitrate controls to Native Flash.

### 5.1 Supported Task Endpoints

- Submit a task: POST /v1/videos

- Retrieve a task: GET /v1/videos/{task_id}

- Typical status values: queued / in_progress / completed / failed / expired

Current task-management scope

Current scope: do not assume support for DELETE /v1/videos/{task_id}, GET /v1/videos, or POST /v1/videos/{task_id}/cancel unless those endpoints are separately enabled and confirmed by Light Speed Future.

### 5.2 Model Families and Output Scope

| Model family | Use case | Supported output | Unsupported by default |
| --- | --- | --- | --- |
| Standard Seedance 2.0 alias | Standard video generation and enabled higher-resolution output | 480p / 720p / 1080p, and 4k when available for your account |  |
| Fast Seedance 2.0 alias | Fast video generation at supported resolutions | Lower supported resolutions such as 480p / 720p | 1080p and 4k |
| Mini Seedance 2.0 alias | Mini video generation at supported resolutions | 480p / 720p, duration 4-15 seconds | 1080p and 4k |
| Seedance 2.5 alias | Multimodal generation, video editing, video extension, and optional Draft Mode | 480p / 720p / 1080p / 4k; text, frame, image, video, audio, and mixed reference modes; Draft Mode: 480p preview → 1080p or 4k Final | 8k |

#### Native bitrate control

For Native Seedance 2.0 Standard, Fast and Mini, and Native Seedance 2.5, set metadata.bitrate_mode to "standard" or "high". If omitted, the default bitrate setting applies. Seedance 2.5 accepts the field in normal, Draft and Final requests with MP4 or MOV output. An explicit Final value may differ from its Draft; omission does not imply inheritance. Native codec selection is not a request field. This Native setting is different from bitrate_level and bitrate in LSF Video Enhancement, LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash.

### 5.3 Native Seedance 2.5 Request Structure

```text
POST /v1/videos
{
  "model": "<YOUR_MODEL_ALIAS>",
  "prompt": "Describe the video you want",
  "metadata": {
    "duration": 4,
    "resolution": "720p",
    "ratio": "adaptive",
    "generate_audio": false,
    "client_request_id": "native-structure-001"
  }
}
```

### 5.4 Top-Level Fields

| Field | Required | Description |
| --- | --- | --- |
| model | Yes | The video model alias supplied by Light Speed Future. Use the exact alias provided for your account. |
| prompt | Conditional | Required for text-only requests. Optional when supported reference or frame inputs are supplied. |
| metadata | Object | Use an object for the documented controls. Seedance 2.5 requests do not support generic user or seed fields. Follow section 5.7. |

### 5.5 metadata Fields

| Field | Type | Description |
| --- | --- | --- |
| draft | boolean | Seedance 2.5 Draft Mode only. true creates a 480p Draft preview. Omitted or false uses the normal native-generation flow. Draft Mode Final does not resend draft=true; it references the completed Draft through metadata.content. See section 5.8. |
| duration | integer | Optional for Seedance 2.5; omission uses the model default (-1). Accepted values are -1 or an integer from 4 to 30. Edit accepts omitted or -1; extend accepts omitted, -1, or 4-30. |
| resolution | string | Output resolution. Seedance 2.5 supports 480p, 720p, 1080p, and 4k. Its ordinary default remains 720p. Mini supports 480p/720p. Standard Seedance 2.0 can use 4k when enabled. Fast and Mini do not support 1080p or 4k. |
| ratio | string | Seedance 2.5 text/reference modes support adaptive, 16:9, 4:3, 1:1, 3:4, 9:16, and 21:9. Edit, extend, first-frame, and first+last-frame modes support adaptive only. Omission uses adaptive. |
| generate_audio | boolean | Whether to generate audio. Seedance 2.5 accepts omitted, true, or false; send an explicit value when deterministic behavior is required. |
| watermark | boolean | Optional Seedance 2.5 watermark switch. |
| client_request_id | string | Optional request ID. Use 1–128 characters from A–Z, a–z, 0–9 and . _ : - . Follow section 4.1 and keep the same key and payload for replay. |
| content | array | Ordered multimodal input items. Each media item uses type for media format and role for purpose; see section 5.6. Limits: 30 reference_image, 10 reference_video, 10 reference_audio, and 50 total. Pure-audio and mixed references are supported. Draft Mode Final is the documented exception and uses one draft_task item as described in section 5.8. |
| output_format | string | Optional output container: "mp4" or "mov". Omitted requests use MP4. |
| return_last_frame | boolean | Requests the generated last frame. On success, metadata.last_frame_url is present only when returned. |
| omni_reference_task_type | string | Optional for Seedance 2.5 Omni reference-to-video. Allowed values: auto, reference, edit, or extend. Omission uses auto behavior. reference requires reference input; edit and extend require reference_video and the rules in section 5.7. |
| bitrate_mode | string | Native output bitrate profile: standard or high. Supported for Seedance 2.0 Standard / Fast / Mini and Seedance 2.5, including 2.5 Draft / Final with MP4 / MOV. Omission stays unspecified. Final may explicitly use a different value from its Draft. |

Use metadata for Seedance-specific controls

Place duration, resolution, ratio and generate_audio under metadata, not at the top level.

### 5.6 metadata.content Item Structure and Roles

metadata.content is an ordered array. Each item requires type (media format) and role (purpose). Supported roles are reference_image, reference_video, reference_audio, first_frame, and last_frame. Use image_url for image/frame roles, video_url for reference_video, and audio_url for reference_audio. Item order is preserved for @Image1, @Video1, and @Audio1. Draft Mode Final is the only documented exception: metadata.content contains exactly one draft_task item without a media role; see section 5.8.

#### Reference image item

```json
{ "type": "image_url", "role": "reference_image",
 "image_url": { "url": "https://example.com/image-1.jpg" } }
```

#### Reference video item

```json
{ "type": "video_url", "role": "reference_video",
 "video_url": { "url": "https://example.com/video-1.mp4" } }
```

#### Reference audio item

```json
{ "type": "audio_url", "role": "reference_audio",
 "audio_url": { "url": "https://example.com/audio-1.mp3" } }
```

Reference video items use type=video_url with role=reference_video. Keep reference media accessible throughout processing; commercial terms are outside this technical guide.

### 5.7 Seedance 2.5 API Request Specification

Seedance 2.5 is available only when Light Speed Future provides a model alias for your account. Use only the documented fields and combinations below.

| Setting | Supported in Seedance 2.5 | Notes |
| --- | --- | --- |
| Model | Exact model alias provided by Light Speed Future | Use only the model alias provided by LSF. |
| Resolution | 480p, 720p, 1080p, or 4k | Omitted resolution defaults to 720p. 4k uses the same Native Seedance 2.5 alias and existing product permissions. |
| Duration | omitted, -1, or integer 4-30 | Omission uses -1. Edit accepts omitted or -1; extend accepts omitted, -1, or 4-30. Do not send strings, decimals, or null. |
| Input mode | Text; frame control; reference images, videos, or audio; pure audio; or mixed references | Prompt is required for text-only requests and optional when supported reference or frame content is supplied. Maximum 30 images, 10 videos, 10 audio items, and 50 total; original item order is preserved. |
| Ratio | adaptive, 16:9, 4:3, 1:1, 3:4, 9:16, or 21:9 | Text-to-video and reference-to-video support all listed values. Editing, extension, and first/last-frame modes support adaptive only. Omission uses adaptive. |
| Generated audio | generate_audio omitted, true, or false | Set an explicit value when deterministic behavior is required. |
| Not supported | Video callbacks; task cancel, delete, or list; native codec selection; inline file upload, Base64, or data: URLs | Reference media is sent for Native Seedance processing. LSF Seedance 2.5 Production also measures reference-video duration for billing; see section 6.5. |
| Output controls | output_format="mp4" or "mov"; return_last_frame boolean; watermark boolean | Omitted output_format uses MP4. metadata.last_frame_url appears only when the service returns a last frame. |
| Omni task type | omitted (auto behavior), auto, reference, edit, or extend | reference requires reference input. edit requires reference_video, ratio omitted or adaptive, and duration omitted or -1. extend requires reference_video and adaptive ratio; duration may be omitted, -1, or 4–30. Supported input-media limits also apply; confirm reference-video requirements with LSF. |
| Draft Mode | 480p Draft preview → 1080p or 4k Final | Set metadata.draft=true for the Draft stage. Finalize a completed Draft before 7 days have elapsed from its creation, using one draft_task content item with its LSF public task ID. Final inherits the creative inputs; bitrate_mode is a stage-specific output control and may be set independently as described in section 5.8. |
| Bitrate mode | standard or high | Optional in normal, Draft and Final requests with MP4 or MOV output. LSF forwards explicit values and leaves omission unspecified. Final may explicitly differ from Draft. |

Seedance 2.5 request specification | Use documented fields only. For Omni editing and extension, keep the explicit task type, prompt intent, reference assets, ratio, and duration mutually consistent.

#### Seedance 2.5 Explicit Omni Edit Example

```text
POST /v1/videos
{
  "model": "<YOUR_SEEDANCE_2_5_MODEL_ALIAS>",
  "prompt": "Edit @Video1 using the visual details from @Image1 and the sound style from @Audio1.",
  "metadata": {
    "client_request_id": "video-job-20260824-edit-0001",
    "omni_reference_task_type": "edit",
    "duration": -1,
    "resolution": "1080p",
    "ratio": "adaptive",
    "output_format": "mp4",
    "return_last_frame": true,
    "watermark": false,
    "generate_audio": false,
    "content": [
      {
        "type": "video_url",
        "role": "reference_video",
        "video_url": {
          "url": "https://example.com/video-1.mp4"
        }
      },
      {
        "type": "image_url",
        "role": "reference_image",
        "image_url": {
          "url": "https://example.com/image-1.jpg"
        }
      },
      {
        "type": "audio_url",
        "role": "reference_audio",
        "audio_url": {
          "url": "https://example.com/audio-1.mp3"
        }
      }
    ]
  }
}
```

Use metadata.omni_reference_task_type only for Seedance 2.5 Omni reference-to-video requests. Omitting it uses auto behavior and lets the model infer the subtask. Explicit reference, edit, or extend validates known task constraints earlier, but the prompt and inputs must still match; otherwise the task can fail asynchronously with InvalidParameter.TaskTypeMismatch. Do not send a generic task_type field.

#### Seedance 2.5 Explicit Omni Extend Example

```text
POST /v1/videos
{
  "model": "<YOUR_SEEDANCE_2_5_MODEL_ALIAS>",
  "prompt": "Continue @Video1 forward, preserving the scene and motion.",
  "metadata": {
    "client_request_id": "video-job-20260824-extend-0001",
    "omni_reference_task_type": "extend",
    "duration": 5,
    "resolution": "480p",
    "ratio": "adaptive",
    "generate_audio": false,
    "content": [
      {
        "type": "video_url",
        "role": "reference_video",
        "video_url": {
          "url": "https://example.com/video-1.mp4"
        }
      }
    ]
  }
}
```

For edit, omit duration or set it to -1. For extend, omit duration or use -1 or an integer from 4 to 30. Both modes require at least one reference_video; omit ratio or set it to adaptive.

#### Native 4K access output and pricing

LSF provides Native Seedance 2.5 4K through its BytePlus whitelist access. Use your existing Native Seedance 2.5 model alias and set metadata.resolution="4k". The same account and product permissions apply; no separate 4K model alias, endpoint or customer application is required.

Existing duration, aspect-ratio and audio options continue to apply. At 16:9, Native 4K output is 3840 × 2160; dimensions vary with the selected ratio. MP4 uses HEVC Main 10 with 10-bit 4:2:0 (yuv420p10le). The documented MOV format uses HEVC Rext with 10-bit 4:4:4 (yuv444p10le). Use compatible players and editors. Native does not expose a codec selector, and 10-bit output alone does not establish HDR delivery.

| Native Seedance 2.5 4K input class | Standard USD per million completion tokens |
| --- | ---: |
| No video input | 5.64 |
| With original video input | 3.48 |

Published Native 4K standard rates are in USD per million completion tokens. Text, images and audio-only references use the no-video-input rate. Requests containing original video input use the video-input rate. Account-specific commercial terms apply where agreed. The initial reservation is an estimate; final billing uses actual completion tokens and the pricing accepted for the task. Confirm your assigned 4K capacity rather than assuming the same limits as lower-resolution generation.

### 5.8 Seedance 2.5 Draft Mode

Draft Mode is an optional two-stage Native Seedance 2.5 workflow for reviewing a low-resolution preview before requesting the final 1080p or 4k result. It uses the same Seedance 2.5 model alias provided by LSF and the same POST /v1/videos and GET /v1/videos/{task_id} endpoints. It is separate from LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash.

#### Stage 1 · Create the 480p Draft

Create the Draft with metadata.draft=true. Draft creation follows the ordinary Seedance 2.5 input and mode rules in sections 5.6–5.7, but the Draft output is 480p only. Omit resolution to use the Draft default or set metadata.resolution="480p" explicitly. Any other Draft resolution is rejected before task creation.

```text
POST /v1/videos
{
  "model": "<YOUR_SEEDANCE_2_5_MODEL_ALIAS>",
  "prompt": "A ceramic cup on a desk in soft daylight.",
  "metadata": {
    "draft": true,
    "duration": 5,
    "resolution": "480p",
    "bitrate_mode": "high",
    "generate_audio": false,
    "client_request_id": "seedance-draft-001"
  }
}
```

Save the returned LSF public task ID and poll that task to completed. A Draft must complete successfully before it can be used for Final. Final may use a valid replacement key issued for the same LSF organization and account, with the required product access permissions. A shared organization name alone does not establish access; contact LSF if the replacement key cannot use an eligible Draft.

#### Stage 2 · Create the Final

Create Final as a new POST with a new client_request_id. Use the same Seedance 2.5 model alias provided by LSF. metadata.content must contain exactly one draft_task item whose id is the completed Draft's LSF public task ID. Final output defaults to 1080p. Omit resolution, set metadata.resolution="1080p", or set metadata.resolution="4k".

```text
POST /v1/videos
{
  "model": "<YOUR_SEEDANCE_2_5_MODEL_ALIAS>",
  "metadata": {
    "resolution": "1080p",
    "bitrate_mode": "standard",
    "content": [
      {
        "type": "draft_task",
        "draft_task": {
          "id": "<YOUR_COMPLETED_DRAFT_TASK_ID>"
        }
      }
    ],
    "output_format": "mp4",
    "return_last_frame": false,
    "watermark": false,
    "client_request_id": "seedance-final-001"
  }
}
```

For a 4K Final, use the same Final request structure and set metadata.resolution="4k". The Final default remains 1080p; the Draft preview remains 480p.

Final inherits the Draft's creative inputs automatically. Do not resend or change:

- prompt or text content;

- reference images, videos or audio;

- duration;

- ratio;

- generate_audio;

- omni_reference_task_type;

- seed.

For Final, the supported stage-specific controls are metadata.resolution (1080p or 4k), bitrate_mode, output_format, return_last_frame, watermark and client_request_id. Do not combine draft_task with other content items.

bitrate_mode can be specified independently on Final and may differ from the Draft value, as shown above. If Final omits bitrate_mode, LSF also leaves the field unspecified; do not assume LSF copies the Draft value.

#### Draft validity, retry and billing

- Create Final before 7 days have elapsed from the completed Draft’s original creation time.

- Draft and Final are separate tasks. Keep both public task IDs and both client_request_id values in your application records.

- If a Draft or Final POST outcome is uncertain, preserve that stage's original client_request_id and payload. Do not create a replacement job merely because the response was lost.

- Draft is billed independently under the applicable Native Seedance 2.5 480p pricing for the original Draft request inputs.

- Final is a new, independently billed task under the applicable Native Seedance 2.5 pricing for its selected resolution (1080p or 4k). It is charged to the key used to submit Final at the terms accepted for that new task; the original Draft charge is unchanged. The Final pricing class depends on whether the original Draft request included video input; the Draft preview itself does not make Final a video-input request.

- If the original Draft request included video input, the original input-video duration from that Draft request is used where applicable; the generated Draft preview is not counted as input-video duration. Each stage has its own initial charge estimate and final charge. Use your LSF quotation and GET /v1/billing/balance; Draft Mode is not a single combined charge.

### 5.9 Common Video Generation Examples

#### Example A: Minimal text-to-video

```text
POST /v1/videos
{
  "model": "<YOUR_MODEL_ALIAS>",
  "prompt": "A person speaking naturally to camera in a clean indoor setting.",
  "metadata": {
    "duration": 4,
    "resolution": "720p",
    "ratio": "adaptive",
    "generate_audio": false
  }
}
```

#### Example B: Seedance Mini high-bitrate text-to-video

```text
POST /v1/videos
{
  "model": "<YOUR_MINI_VIDEO_MODEL_ALIAS>",
  "prompt": "A premium skincare bottle in soft studio light with a slow camera move.",
  "metadata": {
    "duration": 4,
    "resolution": "720p",
    "ratio": "adaptive",
    "bitrate_mode": "high",
    "generate_audio": false
  }
}
```

#### Example C: Single-image reference

```text
POST /v1/videos
{
  "model": "<YOUR_MODEL_ALIAS>",
  "prompt": "Create a product video using Image 1 as the visual reference. Keep the product shape and label consistent.",
  "metadata": {
    "duration": 4,
    "resolution": "720p",
    "ratio": "adaptive",
    "generate_audio": false,
    "content": [
      {
        "type": "image_url",
        "image_url": {
          "url": "https://example.com/product-reference.jpg"
        },
        "role": "reference_image"
      }
    ]
  }
}
```

#### Example D: Reference video input

```text
POST /v1/videos
{
  "model": "<YOUR_MODEL_ALIAS>",
  "prompt": "Use the camera rhythm and motion style from Video 1, while creating a new product-focused short video.",
  "metadata": {
    "duration": 4,
    "resolution": "720p",
    "ratio": "adaptive",
    "generate_audio": false,
    "content": [
      {
        "type": "video_url",
        "video_url": {
          "url": "https://example.com/reference-camera-motion.mp4"
        },
        "role": "reference_video"
      }
    ]
  }
}
```

#### Example E: Seedance 2.0 Standard 4K output when enabled

```text
POST /v1/videos
{
  "model": "<YOUR_STANDARD_VIDEO_MODEL_ALIAS>",
  "prompt": "Generate a premium product commercial with slow cinematic camera movement and high-detail lighting.",
  "metadata": {
    "duration": 5,
    "resolution": "4k",
    "ratio": "16:9",
    "generate_audio": false
  }
}
```

#### Example F: Native Seedance 2.5 4K

Use the same Native Seedance 2.5 alias and account permissions as your other Native 2.5 requests. Four seconds and silent output are example choices, not product-wide restrictions.

```text
POST /v1/videos
{
  "model": "<YOUR_SEEDANCE_2_5_MODEL_ALIAS>",
  "prompt": "A red ceramic cup on a wooden table in soft daylight, with a slow camera move.",
  "metadata": {
    "duration": 4,
    "resolution": "4k",
    "ratio": "16:9",
    "output_format": "mp4",
    "generate_audio": false,
    "client_request_id": "native-4k-001"
  }
}
```

### 5.10 Unsupported Resolution Behavior

```text
POST /v1/videos
{
  "model": "<YOUR_MINI_VIDEO_MODEL_ALIAS>",
  "prompt": "A short product clip.",
  "metadata": {
    "duration": 4,
    "resolution": "1080p",
    "ratio": "adaptive",
    "generate_audio": false
  }
}
```

Expected result: the request is rejected with an invalid_request_error. No video task is created. Fast and Mini also reject 1080p and 4k. Native Seedance 2.5 supports 4k using the existing Native 2.5 model alias and product permissions.

### 5.11 Retrieve Task Status

```text
GET /v1/videos/{task_id}
{
  "id": "task_xxx",
  "task_id": "task_xxx",
  "object": "video",
  "model": "<YOUR_MODEL_ALIAS>",
  "status": "completed",
  "progress": 100,
  "metadata": {
    "url": "https://example.com/generated-video.mp4",
    "last_frame_url": "https://example.com/generated-last-frame.jpg"
  },
  "usage": {
    "completion_tokens": 40594,
    "total_tokens": 40594
  }
}
```

Poll the same GET /v1/videos/{task_id} every 30 seconds or longer. If a connection is interrupted, resume polling the same task ID; do not submit another POST only because polling was interrupted. Stop when status is completed, failed, or expired. Video callbacks are not supported.

#### Seedance 2.5 Error Details

When a task reaches failed or expired, check the error object returned by GET /v1/videos/{task_id}:

```json
{
  "status": "failed",
  "error": {
    "code": "InvalidParameter.TaskTypeMismatch",
    "message": "The task type identified by the model does not match the specified value."
  }
}
```

Use error.code for application logic and error.message for diagnostics. If a specific error code is unavailable, the API returns video_generation_failed.

### 5.12 Idempotency

Use metadata.client_request_id and retain the public task ID. Follow section 4.1 for the current ID format, API-key scope, identical replay and unknown submission outcomes. Native billing continues to use actual returned usage and the applicable account terms; the finished-video pricing formulas for LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash do not apply to Native generation.

## 5A. Native Seedance 2.5 Flash

Use Native Seedance 2.5 Flash for native 480p or 720p output. Your account needs its own enabled Native Flash alias. This chapter covers its request fields, result links and completion-token billing.

### 5A.1 Choose the correct product

Native Seedance 2.5 Flash is a separate native model billed by completion tokens. Use the exact enabled Native Flash alias issued by LSF; existing Native 2.5 and LSF Flash aliases select different products.

| Product | Output and workflow | Billing unit |
| --- | --- | --- |
| Native Seedance 2.5 Flash | Native 480p or 720p; one ordinary generation task; MP4 or MOV | Actual completion tokens under agreed account terms |
| Native Seedance 2.5 — section 5 | Existing native output through 4K, including the separate Draft-to-Final workflow | Actual completion tokens under its own account terms |
| LSF Seedance 2.5 Flash — section 7 | Existing Fast/Standard generation-and-enhancement workflow; finished MP4 | Billable seconds under the section 7 rules |

Existing products keep their options and prices. Native Flash does not use enhancement to produce higher resolutions.

### 5A.2 Request and output scope

Submit `model`, `prompt` and `metadata` to `POST /v1/videos`; retrieve the returned task ID with `GET /v1/videos/{task_id}`. Use the fields for this product; options from other Seedance workflows are not interchangeable.

| Field or capability | Native Flash scope |
| --- | --- |
| Model alias | The separately assigned alias, using the `lsf-seedance-2.5-native-flash-<tenant>` naming pattern; illustrated as `<YOUR_NATIVE_FLASH_MODEL_ALIAS>` |
| `metadata.resolution` | `480p` or `720p`; omission defaults to `720p` |
| `metadata.duration` | Integer 4–30; omit it or use `-1` for automatic duration |
| Output profile | Native 24 fps, 8-bit; no customer codec, bit-depth or FPS selector |
| `metadata.output_format` | `mp4` or `mov`; MP4 is the default |
| Text input | A non-empty prompt is required for text-only generation |
| Frame input | First frame, or first and last frame, using the existing image-item structure; ratio must be `adaptive` or omitted |
| Reference input | Image and audio references, including audio-only reference requests; up to 10 reference images and 5 reference audio items |
| `metadata.ratio` for text/reference input | `adaptive`, `16:9`, `4:3`, `1:1`, `3:4`, `9:16` or `21:9`; omission uses adaptive behaviour |
| Optional booleans | `generate_audio`, `return_last_frame` and `watermark`; explicit `false` stays false, while omission leaves the default unspecified |
| `metadata.omni_reference_task_type` | Omit or use `auto`; use `reference` only with reference input |
| `metadata.client_request_id` | Recommended for identical-request replay with the same key; see section 5A.4 |

At `16:9`, text-to-video dimensions are 854 × 480 (480p) or 1280 × 720 (720p). Inspect the returned file when exact specifications matter; requested duration does not promise an exact frame count.

For image/audio items and role ordering, follow section 5.6 with the limits above. Frame and reference modes cannot be mixed. Authorized media URLs and `asset://` references retain their access and media constraints; request validation does not measure every remote file's duration or guarantee upstream acceptance.

Video-reference requests are unavailable in this release. This includes editing and extension. Do not send `reference_video`, `edit` or `extend` to Native Flash while this restriction remains in place.

Native Flash has no Draft or Draft-to-Final mode. Omit `draft`; a correctly typed `draft: false` is treated as omission. `draft: true`, null, non-boolean values and any `draft_task` reference are rejected. Native Flash also rejects 1080p/2K/4K/6K/8K output and `bitrate_mode`. The existing flagship Native model's Draft, 4K and bitrate options do not transfer to this alias.

### 5A.3 Text-to-video example

Submit the following JSON body to `POST /v1/videos` with your LSF Bearer API key. Replace the model placeholder with the exact enabled Native Flash alias; never place your API key in a browser application.

```json
{
  "model": "<YOUR_NATIVE_FLASH_MODEL_ALIAS>",
  "prompt": "A slow cinematic view of a quiet mountain lake at dawn",
  "metadata": {
    "resolution": "720p",
    "duration": 4,
    "ratio": "16:9",
    "output_format": "mp4",
    "generate_audio": false,
    "client_request_id": "my-native-flash-request-001"
  }
}
```

For a separately intended MOV task, set `output_format` to `mov` and use a new request ID. Changing an accepted request while keeping its ID is not a format-conversion operation. Do not submit another task merely to recover a missing response or interrupted download.

### 5A.4 Retrieve, download and recover

Retain the returned public task ID, original key identity, request ID and unchanged request. Poll the same task every 30 seconds or longer. A completed task exposes its media URL through `metadata.url`; save retained outputs promptly because result access is temporary. `completed`, `failed` and `expired` are terminal Native task states. Use sanitized error details to diagnose a failure; a delayed or uncertain response does not prove failure.

Native Flash also supports authenticated original-file download through `GET /v1/videos/{task_id}/content`. Download it from your backend with an authorized LSF key. Do not forward that key to an external media URL. This path retrieves the existing result; it does not create a new video. Check the actual downloaded media when exact dimensions, format or playback compatibility matter.

For an accepted Native Flash task, exhaustion of an otherwise valid key's allowance does not by itself prevent query or download. Revocation and current access checks still apply: disabled, expired, deleted or unauthorized keys cannot read the task. A valid replacement key must belong to the same LSF organization and account and retain the required model and IP permissions. Use GET with the original task ID; changing keys does not transfer the original charge or make a repeated POST safe.

An accepted identical request may be replayed with the same key and `client_request_id` to recover the existing task. Different parameters under that ID are a conflict. Without an ID, separate submissions do not provide cross-request deduplication. If acceptance is uncertain, preserve the original identity and request, retrieve the original task when known, and contact support if needed. Do not change keys or create a fresh ID as a blind retry. New tasks still require sufficient allowance and current product availability.

### 5A.5 Billing and availability

An initial reservation is an estimate. Final billing uses valid completion-token usage and the account terms accepted for the original task. Repeated status queries and original-file downloads do not create another generation charge. Image or audio reference input is not video input.

The standard price is USD8.40 per million completion tokens for 480p or 720p requests without video input. Text, image and audio-reference requests use this no-video rate. Your agreed account pricing applies. The token unit price is the same at both resolutions; actual token usage can differ. Video-input pricing is not offered in this release.

Native Flash is available through the LSF API for enabled accounts. It is not available in Light Speed Future for ChatGPT. Completion time depends on the request and service load.

## 6. LSF Seedance 2.5 Production

LSF Seedance 2.5 Production creates videos from prompts and supported image, video and audio references. Choose Lite, Standard or Pro to match your output requirements, with resolutions up to 8K and controls for frame rate, bitrate and optional last-frame delivery.

Submit one request, track one public task and receive the finished video. Use the Production model alias for your selected tier. return_last_frame is unsupported for MOV, 6K and 8K delivery; omit it or set it to false. Other supported MP4 profiles require enabled last-frame delivery.

### 6.1 Product selection and outputs

| Tier | Use this alias | Allowed tier-specific fields |
| --- | --- | --- |
| Lite | &lt;YOUR_PRODUCTION_LITE_MODEL_ALIAS&gt; | Common FPS and bitrate controls. No scene, enhance_style or bit_depth. |
| Standard | &lt;YOUR_PRODUCTION_STANDARD_MODEL_ALIAS&gt; | scene; enhance_style; common controls. No bit_depth. |
| Pro | &lt;YOUR_PRODUCTION_PRO_MODEL_ALIAS&gt; | enhance_style; bit_depth; codec on enabled profiles; common controls. No scene. |

Select Lite, Standard or Pro through model, not tool_version. Choose a resolution explicitly; there is no default. Production accepts named resolutions, not resolution_limit.

| Tier | Resolution values | 6K / 8K format |
| --- | --- | --- |
| Lite | 1080p, 2k, 4k | Not supported |
| Standard | 1080p, 2k, 4k, 6k, 8k | H.264, 8-bit, 4:2:0, MP4 |
| Pro | 1080p, 2k, 4k, 6k, 8k | HEVC, 10-bit, 4:4:4, MP4 |

6K uses a 3456-pixel short side; 8K uses 4320. Exact 16:9 produces 6144 × 3456 or 7680 × 4320 respectively. Other source ratios retain their proportions; the longer side is rounded upward to an even pixel. Delivery does not crop or stretch a near-matching ratio into a fixed canvas. For Pro 6K/8K, use codec=h265, bit_depth=10 and output_format=mp4. Standard omits codec and bit_depth. HEVC 10-bit 4:4:4 requires compatible software. Production 6K/8K does not support ProRes or MOV.

### 6.2 Request fields

| Field | Required | Product behaviour |
| --- | --- | --- |
| model | Yes | Exact LSF Seedance 2.5 Production tier alias provided for your account. |
| prompt | Conditional | Required for text-only generation; optional with supported reference or frame content. The same input-mode rules apply to 6K and 8K. |
| metadata.duration | Optional | JSON integer: omit, use -1, or specify 4–30 subject to the input-mode rules. Omission/-1 selects automatic duration and reserves against 30 output seconds. Edit accepts only omission/-1. Final billing uses the delivered duration. |
| metadata.resolution | Yes | Required: 1080p, 2k or 4k; Standard / Pro also support 6k and 8k. See section 6.1. There is no default. |
| metadata.content | For reference / frame input | Ordered image_url, video_url and audio_url items with the roles in section 5.6. Video references can use an accessible HTTPS URL or an authorized asset:// URI. |
| metadata.<br>omni_reference_task_type | Optional | auto, reference, edit or extend with the native Seedance 2.5 mode rules in section 5.7. |
| metadata.ratio | Optional | Text / reference: adaptive, 16:9, 4:3, 1:1, 3:4, 9:16, 21:9. Frame control, edit and extend: adaptive only. |
| metadata.generate_audio | Optional | Boolean; set explicitly when the application requires a specific audio behavior. |
| metadata.watermark | Optional | Boolean watermark request. |
| metadata.fps | Optional | Sets output FPS: a finite JSON number from 15 to 120, including fractional values. Omission defaults to 24 FPS. There is no FPS surcharge for LSF Seedance 2.5 Production. |
| metadata.bitrate_level / bitrate | Optional | Preset low / medium / high, or an integer target bitrate of 10–150000 kbps. Use one of these controls per request. |
| metadata.scene | Standard only | common, ugc, short_series, aigc or old_film. Standard omission is set to common. |
| metadata.enhance_style | Standard / Pro only | hd or natural. |
| metadata.bit_depth | Pro only | Use the bit depth supported by your selected output format. Pro 6K/8K requires 10 with codec=h265 and MP4. Values 8, 12 and 16 are not interchangeable alternatives for that format. Bit depth alone does not establish HDR. |
| metadata.codec | Enabled Pro profile only | String; requires explicit bit_depth. Pro 6K/8K uses h265 with bit_depth=10 and MP4. Standard / Lite omit codec. See section 6.1. |
| metadata.<br>return_last_frame | Optional | Unsupported for MOV, 6K and 8K: omit or set false. For other supported MP4 profiles, true requests a JPG when enabled. This is separate from an input last_frame reference. |
| metadata.output_format | Optional | Must match the enabled codec / bit-depth combination. Use mp4, or omit, for Standard / Pro 6K/8K. A container name alone does not select a codec. |
| metadata.client_request_id | Strongly recommended | Optional; strongly recommended. 1–128 characters from A–Z, a–z, 0–9 and . _ : - . Same-key identical replays return the existing task. Follow section 4.1. |

The only top-level fields are model, prompt and metadata; metadata must be an object. Put generation and output controls inside it. metadata.codec is available for supported Pro output formats and requires bit_depth; see section 6.1. Do not send top-level codec, mode, task_type, tool_version, video_url, bitrate_mode, seed, camera_fixed, draft, draft_task, resolution_limit or callbacks. Reference videos belong in metadata.content for supported profiles, including Standard and Pro 6K/8K. Use the URL and asset rules in section 6.5.

The reference limits are 30 reference images, 10 videos, 10 audio items and 50 total media items. Keep their order. First-frame input may be used alone or with one last frame; a last frame alone, audio-only reference input and mixing frame-control items with reference media are not supported. Frame, edit and extend modes require adaptive ratio. Edit/extend require a reference video; edit requires automatic output duration. Input-media requirements also apply, so use reference media within the limits confirmed by LSF.

### 6.3 Example A · Standard text to 1080p

```text
POST /v1/videos
```

```json
{
  "model": "<YOUR_PRODUCTION_STANDARD_MODEL_ALIAS>",
  "prompt": "A ceramic cup on a desk in soft daylight. The camera is still.",
  "metadata": {
    "duration": 5,
    "resolution": "1080p",
    "fps": 24,
    "scene": "common",
    "generate_audio": false,
    "return_last_frame": false,
    "client_request_id": "production-standard-001"
  }
}
```

A successfully accepted LSF Seedance 2.5 Production request returns HTTP 202. Save its public task identifier and track that task until the finished video is ready. No additional processing request is needed.

```json
{
  "id": "<YOUR_TASK_ID>",
  "task_id": "<YOUR_TASK_ID>",
  "status": "queued"
}
```

This response shows selected fields. Query the same public task with an authorized API key; see section 2 for key rotation. Section 4.2 documents the shared Production and Flash response format. An accepted identical replay returns HTTP 200 and the current task view.

### 6.4 Example B · Lite 2K with an HTTPS video reference

```text
POST /v1/videos
```

```json
{
  "model": "<YOUR_PRODUCTION_LITE_MODEL_ALIAS>",
  "prompt": "Use @Video1 as a motion reference for a new product scene.",
  "metadata": {
    "duration": 5,
    "resolution": "2k",
    "fps": 60,
    "generate_audio": false,
    "content": [
      {
        "type": "video_url",
        "role": "reference_video",
        "video_url": {
          "url": "https://example.com/reference-video.mp4"
        }
      }
    ],
    "return_last_frame": false,
    "client_request_id": "production-lite-reference-001"
  }
}
```

metadata.duration requests the output duration; it is not the input-video duration. LSF validates the input video and measures its duration before calculating the initial charge estimate.

### 6.5 Video URLs, assets and measured input duration

| Input form | How it is handled |
| --- | --- |
| HTTPS video URL | LSF validates the video and measures its duration before acceptance. Supply a direct, stable and accessible media URL. A link that plays in a browser can still fail this validation. |
| asset:// video reference | The asset must be authorized for your account, have video type and Active status, and be accessible to LSF. LSF measures the video’s duration automatically. |
| Multiple video references | Input seconds are accumulated across the submitted video inputs. Fractional seconds are preserved; do not round each reference up. Repeated video inputs can contribute repeatedly. |
| Image or audio references | Do not add input-video seconds. A last-frame still image also remains a no-video-input request unless a video is included. |
| Accepted identical replay | Reuses the input-duration and pricing information already accepted for that task. It does not re-measure accepted inputs or create a new paid job. |
| New request | Requires the referenced media to remain authorized and accessible. Do not assume a previous URL or permission check remains valid indefinitely. |

Provide URLs that remain accessible long enough for validation and processing. Private or loopback destinations, unsafe redirects and unsupported or unstable media access are rejected. Validation may require byte-range access; a server that cannot provide consistent media metadata may be unsuitable.

LSF measures reference-video duration automatically and uses that measurement for billing. Do not send a separate input-video duration field.

### 6.6 Pro 4K and Standard / Pro 6K and 8K examples

#### Pro 4K with a video asset and last frame

```text
POST /v1/videos
```

```json
{
  "model": "<YOUR_PRODUCTION_PRO_MODEL_ALIAS>",
  "prompt": "Use @Video1 as a reference for smooth camera motion and lighting.",
  "metadata": {
    "duration": 5,
    "resolution": "4k",
    "fps": 24,
    "generate_audio": false,
    "content": [
      {
        "type": "video_url",
        "role": "reference_video",
        "video_url": {
          "url": "asset://<YOUR_VIDEO_ASSET_ID>"
        }
      }
    ],
    "return_last_frame": true,
    "client_request_id": "production-pro-asset-001"
  }
}
```

Do not include scene in a Pro request. Use only an asset ID that is authorized and Active. Virtual and real-human asset creation and verification remain distinct workflows; a real-human asset does not bypass its verification requirements.

#### Last-frame behavior

- return_last_frame=true requests a JPG only for a supported MP4 profile with enabled last-frame delivery. MOV, 6K and 8K do not support it; omit the field or set it to false for those outputs.

- An input content item with role=last_frame is a generation constraint. It does not request extraction of the delivered video’s last frame.

- The requested outputs belong to the same public task. Do not submit a separate frame-extraction request; optional last-frame extraction is not billed as a separate customer API job.

- Use the returned last-frame URL only when present on a successful result, and store it promptly if it is needed for subsequent work.

- Keep your own project and source records when reusing a frame in another request; a returned frame URL is not proof of media provenance.

The following examples request 6K or 8K video. Send one body to POST /v1/videos using your assigned Production tier alias. Duration, ratio and FPS can vary within section 6.2. Keep return_last_frame omitted or false for these outputs.

#### Standard 6K example

```json
{
  "model": "<YOUR_PRODUCTION_STANDARD_MODEL_ALIAS>",
  "prompt": "A red paper kite moves across a clear sky.",
  "metadata": {
    "duration": 4,
    "ratio": "16:9",
    "resolution": "6k",
    "fps": 24,
    "generate_audio": true,
    "return_last_frame": false,
    "client_request_id": "production-standard-6k-001"
  }
}
```

#### Pro 6K example

```json
{
  "model": "<YOUR_PRODUCTION_PRO_MODEL_ALIAS>",
  "prompt": "A red paper kite moves across a clear sky.",
  "metadata": {
    "duration": 4,
    "ratio": "16:9",
    "resolution": "6k",
    "fps": 24,
    "generate_audio": true,
    "codec": "h265",
    "bit_depth": 10,
    "output_format": "mp4",
    "return_last_frame": false,
    "client_request_id": "production-pro-6k-001"
  }
}
```

#### Standard 8K example

```json
{
  "model": "<YOUR_PRODUCTION_STANDARD_MODEL_ALIAS>",
  "prompt": "A red paper kite moves across a clear sky.",
  "metadata": {
    "duration": 4,
    "ratio": "16:9",
    "resolution": "8k",
    "fps": 24,
    "generate_audio": true,
    "return_last_frame": false,
    "client_request_id": "production-standard-8k-001"
  }
}
```

#### Pro 8K example

```json
{
  "model": "<YOUR_PRODUCTION_PRO_MODEL_ALIAS>",
  "prompt": "A red paper kite moves across a clear sky.",
  "metadata": {
    "duration": 4,
    "ratio": "16:9",
    "resolution": "8k",
    "fps": 24,
    "generate_audio": true,
    "codec": "h265",
    "bit_depth": 10,
    "output_format": "mp4",
    "return_last_frame": false,
    "client_request_id": "production-pro-8k-001"
  }
}
```

Use the API value h265 for HEVC. Pixel format describes the delivered file: do not send pixel_format or a chroma-subsampling parameter. Production Pro 6K/8K uses 10-bit 4:4:4 and requires a compatible decoder; common browser and hardware playback paths may not support it.

### 6.7 Results and progress

LSF Seedance 2.5 Production continues automatically after acceptance; GET reports progress. Keep polling the original task through nonterminal states. A completed result provides the finished video. Standard and Pro 6K/8K deliver MP4 without a last-frame JPG. Other supported MP4 profiles may return a separately requested last-frame image.

| Response item | Client handling |
| --- | --- |
| id / task_id | Store the public task identifier returned at creation; use it for subsequent GET requests. |
| status | queued and in_progress are nonterminal; completed and failed are terminal. A successful HTTP GET does not itself mean the video succeeded. Do not create a new job because progress is slow. |
| metadata.url | Finished video URL. seconds and metadata.duration, fps and resolution are strings when returned. Save the final response alongside your billing records. |
| metadata.last_frame_url | Optional final-frame image URL. Absent optional fields are omitted, not filled with null. Do not assume this field exists when extraction was not requested or the task has not completed successfully. |
| error | On failure, use the documented error code/message returned by LSF. Do not build client logic around undocumented service details. |

LSF Seedance 2.5 Production does not currently echo client_request_id in this response. Keep your own request-ID to task-ID mapping. Progress values are indicative, not a promised completion schedule. Named resolutions identify output bands and do not guarantee one exact width and height across aspect ratios.

Result URLs are temporary

After completed, download from metadata.url or use the authenticated original-file route below. Download metadata.last_frame_url separately when present. Copy needed files to your own storage; neither route is a permanent archive or a promise of unlimited link renewal.

#### Download the original video

When status is completed, download the file with a key authorized to read the task:

```text
curl --fail --show-error \
  'https://ai-api.lightspeedfuture.com/v1/videos/<YOUR_TASK_ID>/content' \
  --header 'Authorization: Bearer <LSF_API_KEY>' \
  --output result.mp4.part
```

Rename the temporary file after the transfer completes and your integrity checks pass. Keep your API key on your backend; do not forward the LSF Bearer token to an external media URL. If the transfer fails, download the same task result again without creating a new video. Result access is temporary; save files promptly. Resumable downloads are not guaranteed.

#### Output file information

| Optional response field | Meaning |
| --- | --- |
| metadata.requested_encoding | Requested codec, bit depth, file format and result type; not a measurement of the file. |
| metadata.output_metadata | Available details about the output file, such as codec and bit depth. |
| metadata.delivery_contract | Available media specifications and compatibility information. |
| metadata.delivery_contract.media | Dimensions, frame rate, frame count, pixel format and duration when available. |

Allow optional fields to be absent and additional fields to appear. Check the downloaded file in your intended software when exact format compatibility matters.

In the Production preview, “Default encoding” means no explicit encoding was requested. Upstream-reported specifications show only available values and remain unverified against the file; when none are reported, the preview says “Upstream did not provide encoding specifications.” Missing reports do not establish an incorrect codec. Only existing measured delivery information may be presented as measured.

### 6.8 Pricing, initial estimate and final billing

| Output | Base USD per billable second<br>No video input | Base USD per billable second<br>With video input |
| --- | --- | --- |
| Standard 6K H.264 MP4 | 0.69 | 0.552 |
| Pro 6K HEVC 10-bit MP4 | 3.49 | 2.792 |
| Standard 8K H.264 MP4 | 0.99 | 0.792 |
| Pro 8K HEVC 10-bit MP4 | 4.49 | 3.592 |

These are reference base rates. Your effective account rate includes the applicable account pricing terms. With video references, the video-input rate applies to the entire combined duration: measured input-video seconds plus rounded-up reported output seconds. For other resolutions, use the rate agreed for your account.

Use the effective rate agreed for your account, tier, resolution and input mode. It already includes any agreed account adjustment; do not apply that adjustment twice. Account-specific rates may differ between Production and Flash. There is no additional FPS multiplier for this product.

| Input mode | Chargeable duration | Final charge basis |
| --- | --- | --- |
| No video input | Final output duration rounded up to a whole second | Agreed no-video rate × rounded output seconds |
| Any video input | Sum of measured input-video seconds + final output duration rounded up to a whole second | Agreed video-input rate × combined seconds |

A still image, a first/last-frame image or an audio-only reference does not by itself activate video-input pricing. Reference video, editing, extension and mixed inputs containing video do. Use the video-input rate for the same tier and resolution, not a rate from another tier.

#### Worked duration examples — no price quotation

| Request | Calculation | Chargeable seconds |
| --- | --- | --- |
| No video input; final output 5.042 s | ceil(5.042) | 6 |
| One 5.000 s input video; final output 5.017 s | 5.000 + ceil(5.017) | 11 |
| Input videos 2.125 s and 3.000 s; final output 5.042 s | 2.125 + 3.000 + ceil(5.042) | 12.125 |
| One still image only; final output 5.042 s | ceil(5.042), no video-input seconds | 6 |

The initial reservation uses the requested output duration, or 30 output seconds when duration is omitted or -1, plus measured input-video seconds where applicable. A reservation is not a spending cap. Final billing uses the output duration reported in your task result rounded up to a whole second, which can differ from the requested length and the measured video-stream duration.

The effective rate and measured input information accepted for a task remain fixed. Later price or account-setting changes do not reprice it. LSF calculates the final charge once from the delivered result and adds or releases the difference from the reservation. Repeated GET and an accepted identical replay do not duplicate charges. Final account records take precedence over rounded display values.

One LSF Seedance 2.5 Production task has one customer charge, including an enabled, requested last-frame image. A confirmed task failure, including failure to deliver a requested last frame, releases the full reservation with no final customer charge. A timeout or unknown submission outcome is not a confirmed failure; keep the original task ID and contact LSF if its status remains unclear.

#### Duration and playback

Generated duration is a target, not an exact frame-count guarantee. For example, a reported output duration of 4.065 seconds is billed as five seconds. At the Standard 8K base rate without video input, that is USD4.95 before account-specific terms.

Download the video for use in a player or editor that supports its resolution and codec. Pro 6K/8K HEVC 10-bit 4:4:4 is not universally supported by browsers or hardware decoders. Audio timing can differ slightly at the end of a compressed track; check the result when editing to a precise timeline. A 10-bit file does not by itself promise HDR, a particular color space or lossless preservation.

### 6.9 Request rejection and safe recovery

| Condition | What to do |
| --- | --- |
| Product or output combination not enabled | Confirm the supplied alias and account availability with LSF. |
| Wrong tier fields | Remove scene from Pro; remove scene, enhance_style and bit_depth from Lite; do not use tool_version to select a LSF Seedance 2.5 Production tier. |
| Unusable asset / URL / duration | Correct the source or authorization before creating an intentionally new request. Do not submit a manually asserted input duration to bypass validation. |
| Insufficient allowance | Check GET /v1/billing/balance and leave sufficient balance for the final charge. |
| Accepted task remains nonterminal | Keep the original task and request ID; resume GET. Contact LSF with the public ID if progress remains uncertain. |
| Terminal failure | Inspect the error and resulting balance adjustment. Start a new job only deliberately, after addressing the cause. |

## 7. LSF Seedance 2.5 Flash

LSF Seedance 2.5 Flash creates videos from prompts and supported image, video and audio references. Choose Fast or Standard through your assigned model alias. Submit one request to POST /v1/videos, track it with GET /v1/videos/{task_id} and download the finished MP4 from metadata.url. Standard includes 6K H.264 8-bit output; Fast supports outputs through 4K.

Flash delivers the finished video through one task. For applications that need a preview-and-approve step before requesting a final video, use Native Seedance 2.5 Draft Mode in section 5.8.

### 7.1 Product selection and outputs

| Tier | Named output resolutions | Additional controls |
| --- | --- | --- |
| Fast | 240p / 360p / 480p / 540p / 720p / 1080p / 2k / 4k | FPS and bitrate controls |
| Standard | The Fast resolutions plus 6k and 8k | FPS and bitrate; scene and enhance_style |

Use &lt;YOUR_FLASH_FAST_MODEL_ALIAS&gt; or &lt;YOUR_FLASH_STANDARD_MODEL_ALIAS&gt; exactly as supplied by LSF. Availability depends on your account and enabled output combinations. Select Fast or Standard through model; do not send engine or tool_version. LSF Seedance 2.5 Flash has no Professional tier. Named resolutions indicate output bands, not one fixed width and height for every aspect ratio.

Standard 6K delivers H.264 8-bit 4:2:0 MP4. Its short side is 3456 pixels; the proportional longer side is rounded upward to an even pixel. Exact 16:9 produces 6144 × 3456. A different actual source ratio produces different dimensions, without cropping or stretching to a fixed canvas. Fast does not support 6K or 8K. Flash does not expose a Professional encoding menu.

### 7.2 Request fields

#### Generation parameters

Only model, prompt and metadata are allowed at the top level. metadata must be an object. Fields in the table below are inside metadata unless marked top level. Send the documented JSON types; numeric strings and null are not substitutes.

| Field | Type and requirement | Usage |
| --- | --- | --- |
| model | String; required; top level | Assigned Fast or Standard alias. |
| prompt | String; conditional; top level | Required and nonempty for text-only generation. May be omitted with supported media input. |
| client_request_id | String; required | 1–128 characters: A–Z, a–z, 0–9 and . _ : - . See section 4.1. |
| duration | Integer; optional | Omit, -1, or 4–30. Omission/-1 is automatic and reserves for 30 output seconds. edit requires omission/-1. |
| ratio | String; optional | adaptive, 16:9, 4:3, 1:1, 3:4, 9:16 or 21:9 for text/reference inputs. Frame, edit and extend modes require adaptive. |
| omni_reference_task_type | String; optional | auto, reference, edit or extend. reference needs a reference input; edit/extend need reference_video. |
| content | Array; optional | Ordered media items as in section 5.6. Omit for text-only input. Text belongs in prompt, not a content text item. |
| generate_audio | Boolean; optional | Set explicitly when required. Omission leaves the setting unspecified. |
| watermark | Boolean; optional | Set explicitly when required. Omission leaves the setting unspecified. |

Frame control accepts one first_frame alone or one first_frame with one last_frame, using image_url items. A last_frame alone, duplicate frame roles or mixing frame-control items with reference images/video/audio is rejected. Reference inputs allow up to 30 images, 10 videos and 10 audio items, with 50 items total. Preserve their order for @ImageN, @VideoN and @AudioN. Audio-only input is unsupported.

Use HTTPS URLs or authorized asset:// references. URLs must be nonempty, without surrounding whitespace or embedded user credentials. LSF Seedance 2.5 Flash reference inputs do not accept plain HTTP, Base64, data URLs or inline file uploads. Input format and media limits confirmed for your account still apply.

#### Final output parameters

| Field | Type | Usage |
| --- | --- | --- |
| resolution | String; optional | Values in section 7.1. Omission alone defaults to 1080p; empty string and null are invalid. |
| fps | Number; optional | Finite 15–120, including fractions. Sets output FPS. Omit to preserve the original frame rate. |
| output_format | String; optional | mp4 only; omission also delivers MP4. |
| return_last_frame | Boolean; optional | Default false. Unsupported for MOV, 6K and 8K; omit or set false. Other supported MP4 profiles require enabled extraction for true. |
| bitrate_level | String; optional | low, medium or high; available to both tiers. |
| bitrate | Integer; optional | 10–150000 kbps; available to both tiers. Use either bitrate or bitrate_level. |
| scene | String; Standard only | common, ugc, short_series, aigc or old_film. Standard omission uses common. Fast rejects this field. |
| enhance_style | String; Standard only | hd or natural. Omit to leave the style unspecified. Fast rejects this field. |

LSF Seedance 2.5 Flash does not expose codec, bit_depth, bitrate_mode, seed, camera_fixed, resolution_limit, draft, draft_task, tool_version, engine, source_fps or video_url controls. It does not accept callback, priority, queue or service-tier settings. Unknown fields are rejected. The standalone LSF Video Enhancement request is a different schema.

### 7.3 Request examples

#### A. Fast · 1080p

```text
POST /v1/videos
{
  "model": "<YOUR_FLASH_FAST_MODEL_ALIAS>",
  "prompt": "A paper boat floats across a quiet pond.",
  "metadata": {
    "client_request_id": "flash-demo-001",
    "duration": 5,
    "resolution": "1080p",
    "fps": 120,
    "ratio": "9:16",
    "generate_audio": false,
    "return_last_frame": false
  }
}
```

Generate and persist client_request_id before submission. The example requests 5 seconds and explicitly disables generated audio. Omission of duration or -1 selects automatic duration and reserves against 30 output seconds. Omission of generate_audio or watermark does not mean false. Omit fps to preserve the source frame rate.

#### B. Standard · video reference

This example uses an existing video as a reference for a new generated scene. Replace its placeholder URL with a stable HTTPS video source accessible to LSF. Use an already authorized Active video asset with asset:// only when that asset workflow is enabled.

```text
POST /v1/videos
{
  "model": "<YOUR_FLASH_STANDARD_MODEL_ALIAS>",
  "prompt": "Use @Video1 as a motion reference for a new product scene.",
  "metadata": {
    "client_request_id": "flash-reference-001",
    "duration": 5,
    "resolution": "1080p",
    "omni_reference_task_type": "reference",
    "ratio": "9:16",
    "generate_audio": false,
    "content": [
      {
        "type": "video_url",
        "role": "reference_video",
        "video_url": {
          "url": "https://example.com/reference-video.mp4"
        }
      }
    ]
  }
}
```

fps is intentionally omitted, so processing preserves the source frame rate. LSF measures each reference video before acceptance; metadata.duration refers to requested output, not input duration. If a video’s duration cannot be established, the request is rejected instead of treating that video as zero seconds. Repeated references count repeatedly.

#### C. Standard · 6K MP4

```json
{
  "model": "<YOUR_FLASH_STANDARD_MODEL_ALIAS>",
  "prompt": "A red paper kite moves across a clear sky.",
  "metadata": {
    "duration": 4,
    "ratio": "16:9",
    "resolution": "6k",
    "fps": 24,
    "generate_audio": true,
    "return_last_frame": false,
    "client_request_id": "flash-standard-6k-001"
  }
}
```

Use the assigned Standard alias and resolution="6k". Omit codec and bit_depth. Keep return_last_frame omitted or false. FPS is explicitly 24 in this example; omission preserves the generated source frame rate.

### 7.4 Task lifecycle and result

Poll the original public task every 30 seconds or longer until completed or failed. The workflow advances automatically. Section 4.2 documents the response structure and field types; the final video is in metadata.url. Use an authorized key as described in section 2, and retain your own client_request_id to task-ID mapping.

For a supported MP4 profile, enabled and requested last-frame delivery returns metadata.last_frame_url. It is unsupported for MOV, 6K and 8K. Named resolutions follow the dimensions explained in section 7.1. Download the returned media promptly.

### 7.5 Reservation and final billing

Use the effective rate agreed for your account, tier, final resolution and input mode. Video-input rates are configured independently for this product. Account-specific terms may differ between products. No additional FPS multiplier applies.

| Output | Base USD per billable second<br>No video input | Base USD per billable second<br>With video input |
| --- | --- | --- |
| Standard 6K H.264 MP4 | 0.49 | 0.39 |

These are reference base rates. Your effective account rate includes the applicable account pricing terms. Use the agreed rate for other resolutions, including Standard 8K. The video-input rate applies to the entire combined duration below.

| Input mode | Final charge |
| --- | --- |
| No video reference | Agreed no-video rate × final output seconds rounded up to a whole second. |
| One or more video references | Agreed video-input rate × (total measured original input-video seconds + final output seconds rounded up to a whole second). |

Only video references activate video-input pricing. Images and audio do not add video-input seconds. Keep fractional input seconds; round only the output, not each input or the combined total. One Flash task has one customer charge. Input-video seconds include only the video references supplied in your request.

For a 5-second request without video input, the reservation uses 5 output seconds. If the delivered output is 5.025 seconds, final billing uses 6 output seconds. With a measured 5-second reference video, that same delivered result uses 5 + 6 = 11 chargeable seconds at the agreed video-input rate. Requested duration is not a guarantee of exact media duration.

If duration is omitted or -1, the initial reservation uses 30 output seconds, plus any measured video input. The accepted effective rate and input measurements remain fixed for the task; later configuration or price changes do not reprice it. LSF settles once and releases or collects the difference. Repeated GET does not add charges.

A confirmed task failure, including failure to deliver a requested last frame, releases the full reservation with no final customer charge. A network timeout or unknown submission outcome is not a confirmed failure. Keep the original task ID and contact LSF if its status remains unclear.

### 7.6 Errors and recovery

Use the shared submission errors in section 11.1 and the replay rules in section 4.1. Missing or invalid client_request_id is rejected before task acceptance. Reuse the original key, ID and saved request for an uncertain POST; once a task ID is known, use GET. Do not create a replacement task merely because polling is slow or interrupted.

## 8. Asset Library

Build a reusable media library for Native Seedance, LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash. Asset Library operations use your LSF Bearer API key, with access to the libraries and media authorized for your account. Use the supported asset types for your chosen product and complete real-human verification where required.

### 8.1 Important Principles

- Asset Library requests use the Light Speed Future Bearer API key. You do not need to implement separate signing or project-selection logic.

- Asset Library requests do not require a model field. LSF handles authentication and project selection automatically.

- Do not include undocumented fields in Asset Library requests. LSF manages the required service configuration for your account; use only the fields documented in this guide.

- If a request includes project-related fields, Light Speed Future may ignore, override, or reject them.

- CreateAsset is asynchronous. After upload, poll GetAsset or ListAssets. Only assets with Status=Active can be used for video generation.

- CreateAsset supports URL-based upload. Base64 upload is not supported.

- When an asset from the Asset Library is used for video generation, reference it in metadata.content as asset://asset-xxxxx.

### 8.2 Interface Groups

- Virtual asset library: for virtual character / virtual human assets.

- Real-human asset library: for real-human assets. Real-person verification must be completed before the asset can be used.

### 8.3 Virtual Asset Library Paths

- POST /v1/seedance/virtual/asset-groups/create

- POST /v1/seedance/virtual/asset-groups/list

- POST /v1/seedance/virtual/asset-groups/get

- POST /v1/seedance/virtual/asset-groups/update

- POST /v1/seedance/virtual/asset-groups/delete

- POST /v1/seedance/virtual/assets/create

- POST /v1/seedance/virtual/assets/list

- POST /v1/seedance/virtual/assets/get

- POST /v1/seedance/virtual/assets/update

- POST /v1/seedance/virtual/assets/delete

### 8.4 Real-Human Asset Library Paths

- POST /v1/seedance/real-human/validate-session/create

- POST /v1/seedance/real-human/validate-result/get

- POST /v1/seedance/real-human/asset-groups/list

- POST /v1/seedance/real-human/asset-groups/get

- POST /v1/seedance/real-human/asset-groups/update

- POST /v1/seedance/real-human/asset-groups/delete

- POST /v1/seedance/real-human/assets/create

- POST /v1/seedance/real-human/assets/list

- POST /v1/seedance/real-human/assets/get

- POST /v1/seedance/real-human/assets/update

- POST /v1/seedance/real-human/assets/delete

### 8.5 Typical Real-Human Workflow

1. Call validate-session/create to obtain H5Link and BytedToken.

2. Have the end user open the H5Link and complete real-person verification.

3. After verification succeeds, call validate-result/get to obtain GroupId.

4. Call assets/create to upload media under the verified group.

5. Poll GetAsset or ListAssets until Status=Active.

6. Reference the asset in /v1/videos via metadata.content using asset://asset-xxxxx.

### 8.6 Asset Library Examples

#### Example A: Start real-person verification

```text
POST /v1/seedance/real-human/validate-session/create
{
  "CallbackURL": "https://example.com/callback"
}
```

#### Example B: Query real-person verification result

```text
POST /v1/seedance/real-human/validate-result/get
{
  "BytedToken": "<YOUR_VERIFICATION_TOKEN>"
}
```

#### Example C: Upload an asset

```text
POST /v1/seedance/real-human/assets/create
{
  "GroupId": "group-xxxxxxxx",
  "URL": "https://example.com/image.jpg",
  "AssetType": "Image",
  "Name": "real-human-look-1"
}
```

For detailed media requirements, asset status behavior, and related error codes, see the BytePlus documentation linked in section 13 or contact LSF.

### 8.7 Register and read a video asset

Use a virtual asset group already authorized for your account. A real-human asset uses the corresponding verified real-human workflow instead; do not move it into a virtual library to avoid verification.

```text
POST /v1/seedance/virtual/assets/create
{
  "GroupId": "<YOUR_ASSET_GROUP_ID>",
  "URL": "https://example.com/reference-video.mp4",
  "AssetType": "Video",
  "Name": "reference-video"
}
```

Retain the returned asset identifier and query its status before generation:

```text
POST /v1/seedance/virtual/assets/get
{
  "Id": "<YOUR_VIDEO_ASSET_ID>"
}
```

For an authorized real-human video asset, use POST /v1/seedance/real-human/assets/get with its Id. Use the matching library workflow and permissions, rather than assuming all asset paths are interchangeable.

| Asset result item | Meaning |
| --- | --- |
| Id | Asset identifier used after the asset:// scheme. |
| AssetType | Video for LSF Seedance 2.5 Production or LSF Seedance 2.5 Flash reference-video duration measurement. |
| Status | Wait for Active before using the asset in generation. |
| URL | Temporary asset access when returned. It is not a duration measurement or permanent hosting URL. |

LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash can use an authorized asset:// video reference directly. The asset must be Video, Active and available to the requesting account. LSF measures the input duration; clients do not need to fetch a temporary URL and probe it themselves. The standalone LSF Video Enhancement video_url input does not accept asset://.

Real-human rights and media reuse

Complete the required real-person verification and retain the necessary content rights. An enhanced video or extracted frame is not automatically verified as eligible original media for a later real-human verification flow. Check that use case with LSF before relying on it.

## 9. Seed Audio Speech Generation

This endpoint generates speech audio through Light Speed Future's API.

### 9.1 Endpoint and Authentication

| Field | Value |
| --- | --- |
| Endpoint | POST /v1/audio/speech |
| Base URL | https://ai-api.lightspeedfuture.com |
| Authorization | Bearer &lt;LSF_API_KEY&gt; |
| Content-Type | application/json |

### 9.2 Supported Modes

| Mode | Description | Notes |
| --- | --- | --- |
| text_only | Generate speech from text | No reference media required |
| audio_url | Use public HTTPS audio reference | Up to three audio_url references; do not mix with image_url |
| image_url | Use public HTTPS image reference | One image_url reference; do not mix with audio_url |
| metadata.client_request_id | Idempotent replay | Recommended for safe retry |

### 9.3 Model Alias

Model alias

&lt;YOUR_SEED_AUDIO_MODEL_ALIAS&gt;<br>
Light Speed Future provides the model alias during onboarding. Use only the alias provided by LSF.

### 9.4 Text-only Request Example

```json
{
  "model": "<YOUR_SEED_AUDIO_MODEL_ALIAS>",
  "input": "Your text to generate as speech.",
  "instructions": "Read in a calm, natural product-demo tone.",
  "metadata": {
    "client_request_id": "audio-job-20260704-0001"
  }
}
```

Text length and safe retry

input is required unless instructions provides the full text to generate. The combined trimmed input and instructions text, including their separating newline, must be 3000 characters or less. Use the original client_request_id for the same logical replay. If a timeout leaves the outcome unknown, do not immediately create a fresh job.

The speech request body has a 128 KiB limit. This is a JSON request limit, not the size limit of a remotely referenced media file.

### 9.5 audio_url Request Example

```json
{
  "model": "<YOUR_SEED_AUDIO_MODEL_ALIAS>",
  "input": "Use @Audio1 as the reference style and read this short line naturally.",
  "metadata": {
    "client_request_id": "audio-job-20260704-0002",
    "references": [
      {
        "type": "audio_url",
        "url": "https://example.com/reference-audio.mp3"
      }
    ]
  }
}
```

- audio_url references must be public HTTPS URLs, accessible to LSF, and available until your task finishes.

- Up to three audio_url references are supported.

- Do not mix audio_url and image_url.

- Do not use audio_data or base64 audio input.

### 9.6 image_url Request Example

```json
{
  "model": "<YOUR_SEED_AUDIO_MODEL_ALIAS>",
  "input": "Read this short product-demo line in a calm voice.",
  "metadata": {
    "client_request_id": "audio-job-20260704-0003",
    "references": [
      {
        "type": "image_url",
        "url": "https://example.com/reference-image.jpg"
      }
    ]
  }
}
```

- image_url references must be public HTTPS URLs, accessible to LSF, and available until your task finishes.

- One image_url reference is supported.

- Do not mix image_url and audio_url.

- A link must be directly downloadable, without a browser login. A public URL may still fail if LSF cannot download or read the file.

### 9.7 Response Example

```json
{
  "id": "aud_xxx",
  "object": "audio.speech",
  "created": 1780000000,
  "model": "<YOUR_SEED_AUDIO_MODEL_ALIAS>",
  "url": "<TEMPORARY_AUDIO_DOWNLOAD_LINK>",
  "url_expires_at": 1780007200,
  "duration": 4.08,
  "original_duration": 4.08,
  "usage": {
    "type": "seconds",
    "duration": 4.08,
    "original_duration": 4.08
  },
  "metadata": {
    "client_request_id": "audio-job-20260704-0001"
  }
}
```

Download link expiration

The returned download link expires. Download and store the audio if long-term access is required.

### 9.8 Idempotency / Replay Behavior

- Use metadata.client_request_id for safe replay: 1–128 characters from A–Z, a–z, 0–9 and . _ : - . Keep the same API key and original payload.

- Within the result retention window, replay the same logical request with its original ID. A request still in progress can return HTTP 409 idempotency_in_progress; a different request under the same ID returns idempotency_conflict.

- Use a fresh client_request_id for an intentionally new generation attempt after a confirmed failure or changed content. A client-side timeout alone does not confirm that the original generation failed.

- Do not reuse one client_request_id across different business jobs or different generation requests.

Seed Audio supports identical-request replay within a two-hour result window; the media URL may expire earlier. After the result expires, reusing the same ID can return HTTP 410 idempotency_result_expired for up to 24 hours. Later reuse can create a new billable request, so use a new ID for each intentionally new job. These time windows do not apply to video tasks. Repeating a failed request with the same ID does not automatically start a new attempt.

### 9.9 Billing Behavior

- The final charge uses original_duration, the original-duration value reported for the successfully generated audio. Do not substitute duration when the two differ. Video-style whole-second rounding does not apply. Use the agreed audio rate and the final account record.

- Validation failures are not charged.

- If audio is not generated because of a generation failure, reference-media download failure, or timeout, any applicable initial charge hold is released according to service behavior and your commercial terms.

- Use GET /v1/billing/balance to check the balance associated with the API key.

- Commercial pricing, invoices, top-ups, and payment terms are governed by the applicable agreement with Light Speed Future.

### 9.10 Reference URL Requirements

Reference URLs must:

- use HTTPS;

- be public and accessible to LSF;

- be available until your task finishes;

- not require cookies, login, private headers, signed local access, VPN, or private network access;

- avoid redirects, anti-bot checks, or hotlink protection that can block service-side fetching;

- not use localhost, private network URLs, file://, asset://, data: URLs, or base64 data.

Reference URL caveat

A URL that works in a browser may still fail if LSF cannot download or read the file.

### 9.11 Unsupported in Seed Audio

Unsupported in the current Seed Audio customer endpoint:

- voice or custom voice selector fields;

- speaker clone or voice clone controls;

- audio_data;

- image_data;

- base64 audio or image input;

- data: URLs;

- file:// URLs;

- asset:// URLs;

- localhost or private network reference URLs;

- undocumented request fields.

Use documented fields only

Use only the model alias and documented request fields provided by Light Speed Future.

### 9.12 Error Handling

Common causes of failed requests:

- combined input + instructions text exceeds 3000 characters;

- unsupported request field;

- invalid or inaccessible reference URL;

- unsupported media format;

- reference-media download or processing failure;

- timeout;

- idempotency conflict or reuse of client_request_id for a different request.

After a failed request

Fix the cause before starting a new attempt. Use a fresh client_request_id for that new job; preserve the old ID while investigating an unknown outcome.

Example validation error:

```json
{
  "error": {
    "message": "audio_data is not supported for Seed Audio.",
    "type": "invalid_request_error",
    "param": "audio_data",
    "code": "invalid_request_error"
  }
}
```

## 10. LSF Video Enhancement

LSF Video Enhancement improves footage you already have, with output options for playback and professional editing. Choose Lite, Standard or Professional, select your output resolution and format, and submit the source video through POST /v1/videos. Retrieve the result through GET /v1/videos/{task_id}.

### 10.1 Choose an enhancement tier

| Tier | Model selection | Mode-specific controls |
| --- | --- | --- |
| Lite | Use &lt;YOUR_VIDEO_ENHANCEMENT_LITE_MODEL_ALIAS&gt;. Omit tool_version. | Common resolution, FPS and bitrate controls. Omit scene, enhance_style and bit_depth. |
| Standard | Use &lt;YOUR_VIDEO_ENHANCEMENT_MODEL_ALIAS&gt;. Omit tool_version or set standard. | scene and enhance_style supported. No bit_depth. |
| Professional | Use the same Standard / Professional alias and set tool_version to professional. | enhance_style and bit_depth supported; codec on enabled profiles. Omit scene. |

Lite uses its own model alias. Do not send tool_version="fast". Standard and Professional share one model alias.

### Supported enhancement output bands

| Tier | Configured output bands | API selection |
| --- | --- | --- |
| Lite | Up to 720p, 1080p, 2K, 4K | 240p / 360p / 480p / 540p / 720p / 1080p / 2k / 4k |
| Standard | Up to 720p, 1080p, 2K, 4K, 6K, 8K | The Lite values plus 6k and 8k, subject to your account settings. |
| Professional | Up to 720p, 1080p, 2K, 4K, 6K, 8K | The Standard values; tool_version=professional. For 6K ProRes, use the format settings below. |

Available output tiers and resolutions depend on the model alias and options enabled for your account. “Up to 720p” is a pricing label, not an API resolution value; do not send “≤720p” as metadata.resolution.

#### 6K output formats

| Tier | 16:9 output | Encoding | File |
| --- | --- | --- | --- |
| Standard | 6144 × 3456 | H.264, 8-bit, 4:2:0 | MP4 |
| Professional | 6144 × 3456 | ProRes, 10-bit, 4:2:2 | MOV |

6K uses a 3456-pixel short side and preserves the source aspect ratio; the proportional longer side is rounded upward to an even pixel. Exact 16:9 produces 6144 × 3456. Standard omits codec and bit_depth. Professional 6K uses codec=prores, bit_depth=10 and output_format=mov. ProRes needs compatible editing software and can produce larger files than MP4.

#### Source requirements for 6K

Use one directly accessible HTTP(S) video with readable media information, a finite positive duration and a source frame rate greater than zero and no more than 120 FPS. Source dimensions must be at most 1440 pixels on the short side and 2560 on the long side, in landscape or portrait orientation. Supply the actual source duration in metadata.duration. Supported clip length depends on the file and processing requirements.

Set output fps from 15 to 120, including fractional values, or omit it to retain the source frame rate. Select resolution="6k", or a resolution_limit from 2161 to 3456; do not send both. A limit in this range selects the 6K price band. Lower limits use lower resolution bands, while 3457–4320 selects the 8K band.

### 10.2 Request fields

| Field | Required | Usage |
| --- | --- | --- |
| model | Yes | Exact model alias provided for the selected LSF Video Enhancement tier. |
| metadata.video_url | Yes | Public direct HTTP(S) source-video URL; HTTPS recommended. No login, cookies or private network required. asset:// is not supported in metadata.video_url for LSF Video Enhancement. |
| metadata.duration | Yes | Finite positive source duration in seconds. Supply an accurate value for the initial charge estimate; final billing uses the actual enhanced output. |
| metadata.tool_version | Optional | On the Standard / Professional alias, standard is the default or choose professional. On the separate Lite alias, omit this field or use lite; do not use fast. |
| metadata.resolution | One resolution selector | String: 240p, 360p, 480p, 540p, 720p, 1080p, 2k or 4k; Standard / Professional also support 6k and 8k when enabled. Section 10.1 defines the 6k formats. Do not send pricing labels such as lte_720p. |
| metadata.resolution_limit | Alternative selector | Integer short-side pixel limit, mutually exclusive with resolution: 128–2160 for Lite; 128–4320 for Standard / Professional. For example, 720 selects a lower-resolution pricing tier. Use the exact values documented here rather than a label such as lte_720p. |
| metadata.fps | Optional | Finite JSON number from 15 to 120, including fractional values. Omit to preserve source FPS. Omission uses the highest FPS band only for the initial billing estimate. |
| metadata.bitrate_level | Optional | low, medium or high. Omission preserves the service default behavior. |
| metadata.bitrate | Optional | Integer 10–150000 kbps. Use either bitrate or bitrate_level in a request for clear output intent. |
| metadata.scene | Standard only | common, ugc, short_series, aigc or old_film. Set an explicit value when needed; omission leaves the default setting in effect. |
| metadata.enhance_style | Standard / Professional only | hd or natural. |
| metadata.bit_depth | Professional only | Use the bit depth supported by your selected output format. Professional 6K ProRes requires 10 with codec=prores. Values 8, 12 and 16 are not interchangeable alternatives for that format. |
| metadata.codec | Enabled Professional profile only | String; requires explicit bit_depth. Professional 6K uses prores with bit_depth=10. Standard / Lite omit codec. |
| metadata.output_format | Optional | Must match the enabled encoding. Standard 6K uses MP4; Professional 6K ProRes uses mov. See section 10.1. |
| metadata.client_request_id | Recommended | 1–64 characters from A–Z, a–z, 0–9 and . _ : - . Recommended for retries; use the same key, ID and saved payload. See section 4.1. |

Supply exactly one resolution selector. Unknown metadata and fields that do not apply to the selected tier are rejected. The recommendation to request no more than four times the source FPS is quality guidance, not an additional pricing tier.

Resolution and advanced controls

Lite supports outputs through 4K; Standard and Professional include account-enabled higher-resolution profiles. Section 10.1 lists the 6K output formats. metadata.codec and output_format are accepted only with a compatible, enabled encoding combination; codec requires an explicit bit_depth and is Professional-only. Use the documented format settings for the selected resolution. ProRes ignores bitrate and bitrate_level; omit both for ProRes.

Standalone Video Enhancement does not accept metadata.return_last_frame, including false. Omit the field. MOV, 6K and 8K output does not include a requested last-frame image. Supported ProRes 6K does not imply availability of EXR sequences, FFV1 16-bit or HEVC 12-bit profiles.

### 10.3 Request examples

#### A. Standard · 1080p

```text
POST /v1/videos
```

```json
{
  "model": "<YOUR_VIDEO_ENHANCEMENT_MODEL_ALIAS>",
  "metadata": {
    "video_url": "https://example.com/input-video.mp4",
    "duration": 5,
    "tool_version": "standard",
    "scene": "aigc",
    "resolution": "1080p",
    "fps": 24,
    "client_request_id": "enhance-standard-001"
  }
}
```

#### B. Lite · 2K

```text
POST /v1/videos
```

```json
{
  "model": "<YOUR_VIDEO_ENHANCEMENT_LITE_MODEL_ALIAS>",
  "metadata": {
    "video_url": "https://example.com/input-video.mp4",
    "duration": 5,
    "resolution": "2k",
    "fps": 60,
    "client_request_id": "enhance-lite-001"
  }
}
```

#### C. Professional · 4K

```text
POST /v1/videos
```

```json
{
  "model": "<YOUR_VIDEO_ENHANCEMENT_MODEL_ALIAS>",
  "metadata": {
    "video_url": "https://example.com/input-video.mp4",
    "duration": 5,
    "tool_version": "professional",
    "resolution_limit": 2160,
    "fps": 24,
    "enhance_style": "natural",
    "bit_depth": 10,
    "client_request_id": "enhance-pro-001"
  }
}
```

Professional rejects scene. Lite rejects scene, enhance_style and bit_depth. LSF Video Enhancement does not require a generation prompt.

A top-level prompt value (text, empty string or null) may be present but is ignored by LSF Video Enhancement. It does not affect processing or billing. Other unknown top-level fields remain unsupported.

#### Standard 6K MP4 example

```json
{
  "model": "<YOUR_VIDEO_ENHANCEMENT_MODEL_ALIAS>",
  "metadata": {
    "video_url": "https://example.com/source-video.mp4",
    "duration": 10,
    "tool_version": "standard",
    "resolution": "6k",
    "fps": 24,
    "client_request_id": "enhance-standard-6k-001"
  }
}
```

Standard must omit codec and bit_depth. It uses the H.264 8-bit MP4 profile. Do not add a codec field merely to restate that default.

#### Professional 6K ProRes example

```json
{
  "model": "<YOUR_VIDEO_ENHANCEMENT_MODEL_ALIAS>",
  "metadata": {
    "video_url": "https://example.com/source-video.mp4",
    "duration": 10,
    "tool_version": "professional",
    "resolution": "6k",
    "fps": 24,
    "codec": "prores",
    "bit_depth": 10,
    "output_format": "mov",
    "client_request_id": "enhance-pro-6k-001"
  }
}
```

Professional omits scene. ProRes does not use bitrate or bitrate_level; omit those controls. No generation prompt is required. Source audio may be re-encoded or resampled; there is no audio-format or channel-layout request parameter.

### 10.4 Task lifecycle and result

After submission, poll GET /v1/videos/{task_id} every 30 seconds or longer. Continue with the same task ID through queued and in_progress, and stop at completed or failed. If a connection is interrupted, resume GET polling; do not submit another POST only because polling was interrupted.

Typical create response:

```json
{
  "id": "task_xxx",
  "task_id": "task_xxx",
  "object": "video",
  "model": "<YOUR_VIDEO_ENHANCEMENT_MODEL_ALIAS>",
  "status": "queued",
  "progress": 0
}
```

Illustrative completed response; returned values depend on the actual output:

```json
{
  "id": "<YOUR_TASK_ID>",
  "task_id": "<YOUR_TASK_ID>",
  "object": "video",
  "model": "<YOUR_VIDEO_ENHANCEMENT_MODEL_ALIAS>",
  "status": "completed",
  "progress": 100,
  "metadata": {
    "client_request_id": "enhance-standard-001",
    "video_url": "https://example.com/enhanced-video.mp4",
    "duration": 5.084,
    "resolution": "1080p",
    "fps": 24,
    "tool_version": "standard"
  }
}
```

Download from metadata.video_url or use the authenticated original-file route below, and store the result promptly. Access is temporary with no uniform guaranteed retention period. metadata.duration and metadata.fps are JSON numbers; resolution and tool_version are strings. An expiry value may be included when known; an already expired URL may be omitted. A successful HTTP response is not proof that the entire media file downloaded.

#### Download the original video

When status is completed, download the file with a key authorized to read the task:

```text
curl --fail --show-error \
  'https://ai-api.lightspeedfuture.com/v1/videos/<YOUR_TASK_ID>/content' \
  --header 'Authorization: Bearer <LSF_API_KEY>' \
  --output result.mov.part
```

Rename the temporary file after the transfer completes and your integrity checks pass. Keep your API key on your backend; do not forward the LSF Bearer token to an external media URL. If the transfer fails, download the same task result again without creating a new video. Result access is temporary; save files promptly. Resumable downloads are not guaranteed.

#### Output file information

| Optional response field | Meaning |
| --- | --- |
| metadata.requested_encoding | Requested codec, bit depth, file format and result type; not a measurement of the file. |
| metadata.output_metadata | Available details about the output file, such as codec and bit depth. |
| metadata.delivery_contract | Available media specifications and compatibility information. |
| metadata.delivery_contract.media | Dimensions, frame rate, frame count, pixel format and duration when available. |

Allow optional fields to be absent and additional fields to appear. Check the downloaded file in your intended software when exact format compatibility matters.

Use result.mp4.part for MP4 output.

### 10.5 Initial estimate and final billing

LSF Video Enhancement prices may be displayed per minute or per billable second. Convert a per-minute rate to a per-second rate by dividing by 60; do not divide a per-second quote again. Billing rounds the output duration reported in your task result up to a whole second, with no one-minute minimum. Use your agreed account rate. The existing FPS bands are:

| Output FPS band | Multiplier |
| --- | --- |
| Up to 30 | 1 |
| More than 30, up to 60 | 2 |
| More than 60, up to 120 | 4 |

- With an explicit fps, the initial reservation uses the declared source duration rounded up to a whole second and the requested FPS band.

- If fps is omitted, the initial charge estimate uses the highest FPS band. Processing preserves the source frame rate, and the final charge uses the actual output FPS.

- Final charge = agreed base rate per second × rounded-up reported output seconds × actual-output FPS multiplier. If quoted per minute, divide that rate by 60 first. Only the output is charged; source seconds are not added separately. The price agreed when you submit stays fixed; any difference from the initial estimate is charged or returned once. The same FPS bands apply to 6K output.

- For example, 5.084 output seconds become 6 chargeable seconds. If the final output is 24fps, the final FPS multiplier is 1 even when fps omission caused a highest-band initial estimate.

- Repeated GET requests do not create tasks or duplicate charges. Replays do not reprice an already accepted task.

For example, 10 chargeable output seconds at 60 FPS use one sixth of the agreed per-minute base rate, multiplied by 2. The minute unit is a price display convention, not a minimum duration or a change to the rounding rule. Confirmed failed enhancement tasks release their reservation; a network timeout alone does not establish failure.

#### 6K prices

| Output at up to 30 FPS | Base USD per billable second | Equivalent USD per output minute |
| --- | --- | --- |
| Standard 6K H.264 MP4 | 0.052 | 3.12 |
| Professional 6K ProRes 10-bit MOV | 0.520 | 31.20 |

Your agreed account rate applies. Multiply by 2 for output above 30 through 60 FPS, or by 4 above 60 through 120 FPS. There is no one-minute minimum. For example, a reported 2.006-second output is billed as three seconds: Standard 6K at up to 30 FPS costs USD0.156 at the listed base rate, before account-specific terms. Use the applicable account rate for other resolutions.

#### Audio and editing

Audio may be re-encoded or resampled, with a small difference at the end of a compressed track. Retain your original soundtrack and check the downloaded result in your editor when precise timing or an unchanged track is required. ProRes 10-bit output requires compatible software. Bit depth alone does not guarantee HDR, additional source detail or lossless preservation.

### 10.6 Input and error checklist

Use a direct public HTTP(S) source that needs no cookies, private headers or private-network access and remains stable throughout processing. Avoid short expiry, redirect chains, anti-bot checks and hotlink protection. Browser playback alone does not prove processing compatibility. Follow the media limits confirmed during onboarding.

An active enhancement task can cause another submission for the same account user to return HTTP 409 with mediakit_task_in_progress. Poll the existing task and submit the next intended job when it has finished; reusing an ID is not a way to increase concurrency.

## 11. Integration and Error Handling

| Area | Implementation guidance |
| --- | --- |
| Request construction | Use exact model aliases and case-sensitive field names. Put documented controls under metadata. Keep ordered content roles intact. |
| Client records | Store public task ID, client_request_id, selected alias, intended settings, creation time, terminal state and final response. Keep your own business lineage if needed. |
| Authorization | Keep keys server-side. Follow the key-rotation rules in sections 2 and 5.8; a key in the same organization does not automatically have access to every task. |
| Retry | Use the same task ID for GET recovery. An ambiguous submit response is not permission to create a fresh job. See section 4. |
| Media access | Use stable, direct media URLs. Allow enough access time for validation, queueing and processing. Copy temporary results promptly. |
| Billing | Use the selected product’s rules: Native by actual usage; Native Draft and Final as separate tasks; LSF Video Enhancement quoted per minute and charged by output seconds/FPS band; LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash by finished-video chargeable seconds without an FPS surcharge; Seed Audio by original_duration. Use the effective account rate once. |
| Playback | High-resolution or high-bit-depth output may need compatible codecs and players. Verify client playback requirements before selecting advanced controls. |
| Error parsing | Handle the standard error envelope and task status. Do not depend on exact message wording or undocumented service details. |

```json
{
  "error": {
    "message": "metadata.video_url must be a public HTTP or HTTPS URL.",
    "type": "invalid_request_error",
    "param": "metadata.video_url",
    "code": "invalid_request_error"
  }
}
```

An HTTP success response from an asynchronous create endpoint indicates acceptance, not successful video delivery. Always use the terminal task result for delivery and final billing.

### Maintenance, conflicts and task access

| Response / situation | Client action |
| --- | --- |
| HTTP 503 during scheduled maintenance | New video submissions may receive an English JSON maintenance message with Retry-After and Cache-Control: no-store. Respect Retry-After. Keep the original request and client_request_id if the POST outcome is uncertain. |
| Existing tasks during maintenance | Continue GET through the documented HTTPS endpoint. A service switch can briefly interrupt access; resume GET on the same task afterward. |
| HTTP 409: request ID reused with different parameters | Follow section 4.1 for each product’s request-ID rules. LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash return PRODUCTION_REQUEST_CONFLICT for a different request under an accepted same-key ID. New Native records can return idempotency_conflict. Preserve the original payload during recovery. |
| Task access denied after account/access changes | Retain the original task ID and contact LSF. Do not assume that a replacement key or changed access context can read every historical task. |

Parse the HTTP status, structured error when present and Retry-After. Do not match one exact English error sentence: message wording may change while the documented error handling stays the same.

### 11.1 Production and Flash task errors

LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash share the following error codes. The list does not replace authentication, rate-limit or edge-proxy errors. For asynchronous failures, inspect the task status returned by GET as described in section 4.2.

| HTTP | Code | What to check |
| --- | --- | --- |
| 400 | invalid_client_request_id | Required LSF Seedance 2.5 Flash ID is missing or invalid. |
| 400 | PRODUCTION_REQUEST_FIELD_INVALID / PRODUCTION_METADATA_REQUIRED | Top-level fields or metadata structure. |
| 403 | PRODUCTION_MODEL_NOT_AUTHORIZED / VIDEO_TENANT_NOT_AUTHORIZED / PRODUCTION_CONFIGURATION_NOT_READY | Assigned alias, key permissions and product availability. |
| 409 | PRODUCTION_REQUEST_CONFLICT | Same-key ID was used for a different request. |
| 409 | PRODUCTION_RESERVATION_NOT_ACCEPTED | Reservation or acceptance did not complete. This code alone does not identify the cause. |
| 422 | PRODUCTION_REQUEST_INVALID | Parameter combination or enabled output configuration. |
| 422 | PRODUCTION_TIN_UNAVAILABLE | Reference video could not be measured reliably. |
| 404 | PRODUCTION_TASK_NOT_FOUND | Task ID, original account ownership and current key permissions. |
| 503 | PRODUCTION_ADMISSION_DISABLED / PRODUCTION_RECOVERY_UNAVAILABLE / PRODUCTION_STATE_UNAVAILABLE | Temporary submission, recovery or state availability. Preserve the original request. |

For example, missing client_request_id on LSF Seedance 2.5 Flash returns HTTP 400 before a new task is accepted:

```json
{
  "error": {
    "type": "production_error",
    "code": "invalid_client_request_id",
    "message": "invalid_client_request_id"
  }
}
```

A rejected request may have no task ID. Retain its HTTP status, structured error, timestamp, original request ID and any request identifier in the response. Contact LSF with these details and a redacted request if necessary. Do not change IDs or keys automatically after a timeout or temporary error, because the original submission may already have been accepted.

## 12. Frequently Asked Questions

### 12.1 How should I choose Native, Production or Flash?

Choose Native for native Seedance controls and optional Draft-to-Final previews; Production for finished video with flexible delivery controls; and Flash for fast video generation. Use the output options listed for each product. A Production or Flash output resolution is not a claim that the Native model supports the same resolution.

### 12.2 How is LSF Seedance 2.5 Production different from LSF Video Enhancement?

Use LSF Video Enhancement to improve a video you already have. Use LSF Seedance 2.5 Production to create a new finished video from a prompt and supported references. Both use the public video task endpoints, with product-specific request fields.

### 12.3 Do I submit tool_version for every product?

Select LSF Seedance 2.5 Production Lite / Standard / Pro and LSF Seedance 2.5 Flash Fast / Standard using the alias supplied by LSF. Do not send tool_version or engine to either product. LSF Video Enhancement Standard / Professional use one alias with tool_version; Lite uses a separate alias.

### 12.4 Does a still image count as video-input seconds?

No. Only video inputs contribute input-video seconds. A last-frame still image is different from a short tail-video reference.

### 12.5 Why can the final charge differ from the initial estimate?

For Production and Flash, the delivered video can be longer than requested. Final output seconds are rounded up, and any measured reference-video seconds are added without whole-second rounding. The accepted effective rate remains fixed. Native uses its separate usage-based billing rules; see the relevant product chapter.

### 12.6 Does omitted fps force 120fps?

No. Production defaults to 24 FPS when fps is omitted. For Video Enhancement, omission retains the source FPS; the initial estimate uses the highest FPS price band and the final charge uses actual output FPS. Production and Flash have no additional FPS surcharge. See each product chapter for its output controls.

### 12.7 Does return_last_frame request a first/last-frame generation constraint?

return_last_frame requests an image of the finished video, not a first/last-frame generation constraint. Production and Flash do not support it for MOV, 6K or 8K; omit it or set false. Other supported MP4 profiles require enabled extraction. Standalone Video Enhancement does not accept this field at all. Native Seedance has its separate controls in section 5.

### 12.8 Will repeated polling advance or charge the workflow again?

No. LSF Seedance 2.5 Production and LSF Seedance 2.5 Flash advance in the background. Repeated GET reads the existing task and does not create a new generation or duplicate billing.

### 12.9 Can I use either asset library without verification?

No. Use the authorized library path for the asset. Real-human assets still require the corresponding verification and rights.

### 12.10 Can I cancel, list or delete video tasks through this guide?

No. The documented video endpoints support creation and retrieval. Asset Library list/delete paths are separate asset operations.

### 12.11 Can I expose the API key in the website?

No. Keep it on your backend. The website may display example placeholders and product information; it should not embed customer credentials.

### 12.12 Can I request MOV from LSF Seedance 2.5 Production or LSF Seedance 2.5 Flash?

Production Standard and Pro 6K/8K deliver MP4, as do the documented Flash outputs. Standalone Video Enhancement Professional 6K supports ProRes 10-bit MOV with its own request schema. Do not transfer that format to Production. Native Seedance has its separate output_format control in section 5.

### 12.13 Can I request another account's asset or task?

Only your authorized assets and task context are available. Knowing an identifier is not authorization.

### 12.14 How do I access a product or output option?

Use the model alias and product permissions assigned to your account. Contact LSF to add a product or output option. The supported settings are listed in the corresponding product chapter.

### 12.15 What is Seedance 2.5 Draft Mode?

It is a two-stage Native Seedance 2.5 workflow: first create and review a 480p Draft, then create a separate 1080p Final, or a 4k Final, from that completed Draft. It uses the same Seedance 2.5 model alias supplied by LSF and the normal video task endpoints.

### 12.16 Is the Final step included in the Draft charge?

No. Draft and Final are independently billed tasks. Draft uses the applicable Native Seedance 2.5 480p pricing for its original request inputs. Final uses the applicable 1080p or 4k pricing class determined by whether the original Draft request included video input; the generated Draft preview itself does not force video-input pricing or add its duration as input video.

### 12.17 Can Final use a different API key, model alias or creative inputs?

A valid replacement key for the same LSF organization and account may create Final from an eligible Draft, subject to current product access permissions. Use the same Seedance 2.5 model alias. Final inherits the Draft’s prompt, references, duration, ratio, audio choice and Omni task type; do not resend or change them. Set supported Final output controls as described in section 5.8. Final is a new task charged to the submitting key; the Draft charge is unchanged.

### 12.18 How long can I keep a Draft before creating Final?

Create Final after the Draft completes successfully and before 7 days have elapsed from its original creation time.

### 12.19 How do I select Standard or High bitrate for Native Seedance?

Set metadata.bitrate_mode to "standard" or "high" on a supported Native request. It is available for the Seedance 2.0 series and Seedance 2.5, including 2.5 Draft / Final with MP4 / MOV. Omission leaves the field unspecified; set it explicitly if needed. An omitted Final value is not assumed to inherit the Draft value.

### 12.20 Can I force Native Seedance to output H.264 or H.265?

No. Native Seedance codec selection is not exposed as a customer request control. Do not send metadata.codec. The delivered codec and bit depth depend on the native model, resolution and output profile; verify playback compatibility for your target environment.

### 12.21 Which product supports a reviewable preview before the final video?

Use Native Seedance 2.5 Draft Mode for a reviewable preview followed by a separately requested Final. Flash delivers a finished video through one task and does not offer a separate preview step.

### 12.22 Does LSF Seedance 2.5 Flash require client_request_id?

Yes. Generate and save it before the first POST. Use a different ID for an intentionally new job and retain the original key, ID and body for retries. Do not reuse one video ID across products on the same key.

### 12.23 Can I choose a codec for LSF Seedance 2.5 Flash or LSF Seedance 2.5 Production?

Flash does not expose a codec selector. Production Pro 6K/8K uses metadata.codec="h265" with bit_depth=10 and MP4; Standard omits codec and bit_depth. Section 6.1 lists Production formats and playback requirements.

### 12.24 Does a named 1080p output always have exactly 1920 by 1080 pixels?

It identifies an output resolution band. Actual width and height depend on the source, aspect ratio and processing profile. For example, a portrait result can have a height slightly different from 1920. Validate the returned media for strict downstream pixel requirements.

### 12.25 Are quoted minutes a one-minute minimum for LSF Video Enhancement?

No. A per-minute rate is a display unit. Billing uses the reported output duration rounded up to whole seconds and the applicable FPS band. A per-second quote is already in the billing unit; do not divide it by 60 again.

### 12.26 Why can two accounts or two products have different effective prices?

Prices can reflect account-specific commercial terms. Use the agreed effective rate for the selected product, tier, resolution and input mode. These terms are applied by LSF, not chosen through a request parameter. Accepted tasks retain their original pricing terms.

### 12.27 How do I request Native Seedance 2.5 4K?

Use your existing Native Seedance 2.5 alias and the ordinary POST /v1/videos endpoint with metadata.resolution="4k". The same Native Seedance 2.5 account permissions apply. Draft previews remain 480p; a Final can request 4k using the completed Draft. No separate 4K alias or customer application is needed.

### 12.28 Does a lower per-token rate make 4K cheaper than 1080p?

Not necessarily. Native 4K uses USD5.64 per million completion tokens without video input and USD3.48 with original video input at the published standard rates. Higher resolution generally uses more tokens. Compare the total expected token usage and your applicable rate, rather than the unit price alone. Actual task usage determines the final charge.

### 12.29 Is Native Seedance 2.5 Flash the same as LSF Seedance 2.5 Flash?

No. Native Flash is the separate 480p/720p native model in section 5A, with completion-token billing and no Draft-to-Final or higher-resolution output. The existing LSF Flash in section 7 retains its Fast/Standard workflow, output options and per-second billing. Use the separately enabled Native Flash API alias. Native Flash is not available in Light Speed Future for ChatGPT.

### 12.30 Can I read an accepted Native Flash task after my key allowance is exhausted?

Yes, provided the key is still valid and authorized for that task. Query and original-file download do not require new generation allowance. Revoked, disabled, expired, deleted, model-restricted or IP-disallowed credentials remain subject to access checks. Use an authorized same-account, same-organization replacement key if needed, retain the original task ID, and do not repeat the POST with a new key. New task creation still requires sufficient allowance and an enabled product.

## 13. Further Reading

This guide and the onboarding information supplied by LSF describe the supported LSF API requests. BytePlus documentation is supplementary for media and asset requirements and does not add extra fields to an LSF request.

| Reference | Use |
| --- | --- |
| LSF website · https://lightspeedfuture.com | Product information and contact. |
| BytePlus ModelArk documentation · https://docs.byteplus.com/en/docs/ModelArk | Generation media guidance, Asset Library and verification requirements. |
| GetAsset · https://docs.byteplus.com/en/docs/ModelArk/2318274 | Asset status and media requirements. Use LSF’s documented endpoints and authentication. |
| LSF Video Enhancement · section 10 | Supported LSF enhancement controls, output selection and billing. Contact LSF for account-specific media requirements. |

Use only the request fields and authentication described in this guide.

## 14. Support

For onboarding, model aliases, enabled outputs, asset permissions, pricing or unresolved task outcomes, contact Light Speed Future.

| Contact | Address |
| --- | --- |
| Email | info@lightspeedfuture.com |
| Telegram | @HenryBroG |
| Website | https://lightspeedfuture.com |

For a task issue, provide the public task ID, client_request_id, product alias, approximate timestamp with timezone and returned error code/message. If the request was rejected before a task ID was created, provide its HTTP status, response body and any request identifier present in the response, plus a redacted request body if needed. Do not send API keys, login cookies or signed media URLs in an ordinary support message.

LSF supplies API keys and model aliases during account setup. Include the relevant product and output requirements when requesting integration assistance.

## Appendix A. Customer Integration Checklist

Complete this sheet for each customer delivery. Send the API key separately through the agreed private channel.

| Item | Account value |
| --- | --- |
| Customer / environment | &lt;ACCOUNT_NAME&gt; / &lt;ENVIRONMENT&gt; |
| API guide version | &lt;GUIDE_VERSION&gt; |
| Base URL | &lt;YOUR_API_BASE_URL&gt; |
| API key | &lt;SENT_SEPARATELY&gt; |
| Native model aliases | &lt;YOUR_SEEDANCE_2_5_MODEL_ALIAS&gt; and any contracted 2.0 aliases |
| Enhancement Standard / Professional | &lt;YOUR_VIDEO_ENHANCEMENT_MODEL_ALIAS&gt; |
| Enhancement Lite | &lt;YOUR_VIDEO_ENHANCEMENT_LITE_MODEL_ALIAS&gt; or not enabled |
| LSF Seedance 2.5 Production Lite / Standard / Pro | &lt;EXACT_ENABLED_PRODUCTION_ALIASES&gt; |
| LSF Seedance 2.5 Flash Fast / Standard | &lt;EXACT_ENABLED_FLASH_ALIASES&gt; |
| Available combinations | &lt;PRODUCT, TIER, RESOLUTION, CODEC, BIT DEPTH, FORMAT AND ACCESS SCOPE&gt; |
| Client request IDs | &lt;USE THE PRODUCT-SPECIFIC REQUEST ID RULES IN SECTION 4.1&gt; |
| Generation inputs | &lt;TEXT / IMAGE / HTTPS VIDEO / ASSET VIDEO / AUDIO / MODE SCOPE&gt; |
| Last-frame delivery | &lt;ENABLED / NOT ENABLED&gt; |
| Asset Library scope | &lt;VIRTUAL / REAL-HUMAN / BOTH / NONE&gt; |
| Speech alias | &lt;YOUR_SEED_AUDIO_MODEL_ALIAS&gt; or not enabled |
| Commercial terms | &lt;LSF_QUOTATION_REFERENCE&gt; |
| Client media requirements | &lt;CODEC, PLAYBACK, OUTPUT RETENTION AND APPROVED MEDIA LIMITS&gt; |
| Support | Telegram @HenryBroG · info@lightspeedfuture.com |
| Native Seedance 2.5 Flash | &lt;YOUR_ENABLED_NATIVE_FLASH_ALIAS_AND_AGREED_ACCOUNT_RATE&gt; |
| Native Seedance 2.5 4K pricing | &lt;CONFIRMED ACCOUNT RATE&gt; |
| Native Seedance 2.5 Draft Mode | &lt;ENABLED / NOT ENABLED; 480p DRAFT → 1080p OR 4k FINAL&gt; |

