Videos

POST /v1/videos/generations is always asynchronous. It returns 202 Accepted with a generation in queued status right away; poll it or receive a webhook when it finishes.

curl https://api.geekidea.ai/v1/videos/generations \
  -H "Authorization: Bearer $GEEKIDEA_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: order-1042-video" \
  -d '{
    "model": "bytedance/seedance-2.5",
    "prompt": "Aerial shot of a futuristic city at dawn",
    "duration": 5,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "generation_config": { "generate_audio": true },
    "metadata": { "order_id": "1042" }
  }'

202 Accepted

{
  "id": "video_0192f1c3a7b87c3e9a4f5b6c7d8e9f01",
  "object": "video",
  "model": "bytedance/seedance-2.5",
  "status": "queued",
  "created_at": 1791425100,
  "estimated_cost": 1.3289,
  "currency": "USD",
  "billing_status": "estimated",
  "metadata": { "order_id": "1042" }
}

Parameters

Field	Description
`model`	A model whose `type` is `video`
`prompt`	What to generate
`duration`	Seconds. Defaults to the model’s default, or 5 seconds when allowed
`resolution`	For example `480p`, `720p`, `1080p`. Defaults to 720p when available
`aspect_ratio`	For example `16:9`, `9:16`, `1:1`. Defaults to the model’s default
`input`	Frames and reference media, see below
`generation_config`	`generate_audio`, `seed`, `negative_prompt`, as the model allows
`metadata`	Up to 16 string key–value pairs, returned with the result

Allowed values come from the model’s generation_config. A value outside them returns 400 unsupported_value, and the message lists the allowed values.

Frames and references

"input": {
  "first_frame": { "image_url": "https://example.com/first.png" },
  "last_frame": { "file_id": "file-0192f1c3a7b87c3e9a4f5b6c7d8e9f01" },
  "reference_images": [{ "image_url": "data:image/png;base64,..." }],
  "reference_video_urls": ["https://example.com/motion.mp4"],
  "reference_audio_urls": ["https://example.com/voice.mp3"]
}

Images take either image_url (https or data URL) or file_id. Videos and audio take https links. What a model accepts is declared in its capabilities (first_frame, last_frame, reference_image, reference_video, reference_audio) and limits (max_reference_images, max_reference_videos, max_reference_audios).

Error	When
`400 input_not_supported`	The model does not accept this kind of input
`400 too_many_inputs`	More references than the model allows
`400 missing_parameter`	A last frame without a first frame, or no image for a model that requires one
`400 conflicting_inputs`	Frames and references together on a model that accepts only one or the other

Some models take the aspect ratio from the first frame or reference image, in which case aspect_ratio is ignored. The model page shows this.

Result

When status becomes completed, the generation’s output.videos contains:

"output": {
  "videos": [
    {
      "url": "https://….mp4",
      "thumbnail_url": "https://….webp",
      "previews": { "preview": "https://…", "lite": "https://…", "grid": "https://…", "icon": "https://…" },
      "seconds": 5.0,
      "width": 1280,
      "height": 720
    }
  ]
},
"usage": { "video_seconds": 5, "cost": 1.3289, "currency": "USD", "cost_details": { "video": 1.3289 } }

Links are signed and expire. Query the generation again to get fresh links.

Pricing

Price per second × the requested duration, by resolution and whether audio is generated (second_per_spec, for example 720p|audio). Failed videos are not charged; for cancellations see below. Seedance models are billed per video token instead (video_tokens_per_spec):

Tokens = actual output width × height × frames / 1024, rounded down, with frames = 24 × seconds + 1. The size is the size of the video you get back (for example 720p is 1280×720 at 16:9 and 834×1112 at 3:4), so every aspect ratio costs about the same.
With reference videos, each reference video adds width × height × 24 × seconds / 1024 tokens, and all tokens are billed at video_tokens_video_input_per_spec.
usage.video_tokens reports the billed tokens. The hold at submission is a slightly larger estimate; you are charged for the actual tokens.

For example, a 4-second 720p 9:16 Seedance 2.0 video (720×1280, 97 frames) is 87,300 tokens.

Cancel

POST /v1/generations/{id}/cancel cancels a video. A queued video is cancelled right away and not charged. A video already being generated is charged only if it can’t be stopped and completes anyway. See Async requests.

Get started

Generate

Workflows

Async and webhooks

Account

Parameters

Frames and references

Result

Pricing

Cancel

​Parameters

​Frames and references

​Result

​Pricing

​Cancel

Parameters

Frames and references

Result

Pricing

Cancel