API

Example request

Set your FLUX_API_KEY before running. See the parameters below for supported settings.

curl https://api.flux-context.org/api/v1/generations \
  -H "Authorization: Bearer $FLUX_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
  "model": "seedance-2-5",
  "input": {
    "mode": "text-to-video",
    "prompt": "A calm seaside sunrise with gentle camera movement",
    "resolution": "480p",
    "duration": 4,
    "sound": false,
    "aspect_ratio": "16:9"
  }
}'

Parameters

Set model to seedance-2-5. These fields belong inside input.

Reference media

Use your own publicly accessible HTTPS media URLs. Links must remain valid throughout queueing and generation. Base64 and file upload bodies are not supported by this endpoint.

curl https://api.flux-context.org/api/v1/generations \
  -H "Authorization: Bearer $FLUX_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
  "model": "seedance-2-5",
  "input": {
    "mode": "first_and_last_frames",
    "prompt": "A calm seaside sunrise with gentle camera movement",
    "resolution": "480p",
    "duration": 4,
    "sound": false,
    "aspect_ratio": "auto",
    "image_urls": [
      "https://example.com/front.jpg",
      "https://example.com/back.jpg"
    ]
  }
}'

curl https://api.flux-context.org/api/v1/generations \
  -H "Authorization: Bearer $FLUX_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
  "model": "seedance-2-5",
  "input": {
    "mode": "omni_reference",
    "prompt": "A calm seaside sunrise with gentle camera movement",
    "resolution": "480p",
    "duration": 4,
    "sound": false,
    "aspect_ratio": "auto",
    "image_urls": [
      "https://example.com/front.jpg",
      "https://example.com/back.jpg"
    ]
  }
}'

curl https://api.flux-context.org/api/v1/generations \
  -H "Authorization: Bearer $FLUX_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
  "model": "seedance-2-5",
  "input": {
    "mode": "omni_reference",
    "prompt": "A calm seaside sunrise with gentle camera movement",
    "resolution": "480p",
    "duration": 4,
    "sound": false,
    "aspect_ratio": "auto",
    "image_urls": [
      "https://example.com/front.jpg"
    ],
    "video_urls": [
      "https://example.com/reference.mp4"
    ],
    "audio_urls": [
      "https://example.com/reference.mp3"
    ]
  }
}'

curl https://api.flux-context.org/api/v1/generations \
  -H "Authorization: Bearer $FLUX_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
  "model": "seedance-2-5",
  "input": {
    "mode": "omni_reference",
    "prompt": "Edit the video to use a warm sunset background",
    "resolution": "480p",
    "duration": -1,
    "sound": false,
    "aspect_ratio": "auto",
    "image_urls": [
      "https://example.com/front.jpg"
    ],
    "video_urls": [
      "https://example.com/reference.mp4"
    ],
    "audio_urls": [
      "https://example.com/reference.mp3"
    ],
    "reference_intent": "edit"
  }
}'

curl https://api.flux-context.org/api/v1/generations \
  -H "Authorization: Bearer $FLUX_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
  "model": "seedance-2-5",
  "input": {
    "mode": "text-to-video",
    "prompt": "A calm seaside sunrise with gentle camera movement",
    "resolution": "480p",
    "duration": 4,
    "sound": false,
    "aspect_ratio": "16:9",
    "draft": true
  }
}'

curl https://api.flux-context.org/api/v1/generations \
  -H "Authorization: Bearer $FLUX_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
  "model": "seedance-2-5",
  "input": {
    "draft_task_id": "00000000-0000-4000-8000-000000000000"
  }
}'

Credits

Video reference seconds are included in the billable duration. See the model-specific rounding and editing rules below. Sound does not change the price. Confirmed failed requests are refunded.

Without video references, credits = ceil(output seconds × rate). With video references, new tasks cost ceil((output seconds + max(measured total input video seconds, ceil(output seconds × 2/3))) × rate). Images and audio add no billable seconds.

Editing reserves ceil((max(input video seconds, 20) + 30 + 1) × rate). The extra second is a Flux settlement reserve, not provider output. Final cost uses ceil(final total billable seconds × the saved rate); the difference is returned or charged. Historical tasks retain their saved rule. If final billing is unavailable, the precharge is retained and results are delivered; billing is marked for review. If the available balance cannot cover an additional charge, only the available credits are charged and results are delivered.

Constraints

Set draft: true for a 480p preview at the normal 480p price. Then submit only draft_task_id (the Flux generation id, not provider_task_id) to create a separately billed 1080p final. The source must be your successful API preview, created within 7 days. Inherited fields must be omitted. Each new Idempotency-Key creates independent paid work.

Normal generation and preview inputs require a prompt and retain the ordinary mode rules. A final request inherits prompt, references, duration and sound from its source; do not send those fields.

Preview: 15 credits/second without video references, 9 with video references. Final: 82 credits/second without video references, 50 with video references. New video-reference tasks bill output seconds + max(actual total reference seconds, ceil(output seconds × 2/3)); apply the minimum once and round total credits up. A new final uses the original actual reference duration, even from an older preview. Preview charges are not deducted from final charges. A dynamic source needs authoritative final billing; DRAFT_BILLING_PENDING means check again later.

Responses include workflow.stage (draft or final), workflow.sourceId for a final, workflow.expiresAt for a preview, and workflow.can_finalize. A website creation cannot be used as an API source. A source is reusable for multiple independent paid finals; retries of one operation must reuse its Idempotency-Key.

First/last frames use 1–2 images and aspect_ratio auto. Omni reference supports up to 30 images, 10 videos and 10 audio files.

Each video/audio reference must be 2–30 seconds; video total and audio total must each be at most 30 seconds.

For explicit editing use reference_intent edit, duration -1 and aspect_ratio auto. Other requests use a fixed duration of 4–30 seconds.

Use returned dimensions when available. Unverified video dimensions are null. Videos are delivered as MP4. Reference and editing rules depend on the model.

Response

A successful submission returns HTTP 202 and a request ID. Response excerpt:

{
  "id": "REQUEST_ID",
  "model": "seedance-2-5",
  "status": "processing"
}