UlazAI docs

Image and video APIs

Video Studio API

Pricing Dashboard Image studio

POST /api/v1/video-studio/generate/

Canonical video generation endpoint

Use Video Studio for all current video models. This endpoint supports text-to-video, image-to-video, and model-specific workflows such as video remix.

For dynamic integrations, fetch model capabilities from /api/v1/models/video/ before sending generation payloads.

Recent Video Studio routes include seedance_2_5, seedance_2_mini, kling_v3_turbo_t2v, kling_v3_turbo_i2v, video_lip_sync, omnihuman_1_5, happyhorse_t2v, happyhorse_i2v, happyhorse_r2v, and happyhorse_videoedit.

Authentication: send Authorization: Bearer YOUR_API_KEY on every request.

Poll with /api/v1/video-studio/status/{job_id}/ and load recent jobs from /api/v1/video-studio/history/.

Request contract

Field	Type	Required	Notes
`model_slug`	string	Yes	Must match a registered Video Studio model slug
`prompt`	string	Usually	Required for most models. Storyboard mode can start with images only.
`aspect_ratio`	string	No	Must be one of the selected model's aspect ratios
`duration_seconds`	integer	No	Must match an allowed duration for the model. Motion Control uses the motion clip length for pricing.
`quality_mode`	string	No	Required when model defines quality modes
`input_image_urls`	array[string]	Model-specific	Used for image-to-video models. Some models require at least one URL.
`input_video_urls`	array[string]	Model-specific	Used by video remix and source-video routes such as `wan_2_6_v2v`, `kling_3_0_motion_control`, `video_lip_sync`, and `happyhorse_videoedit`.
`seedance_mode`	string	Seedance 2 family	Use `text_to_video`, `first_frame`, `first_last_frame`, or `multimodal`.
`first_frame_url`, `last_frame_url`	string	Seedance 2 family	Use for the first-frame and first+last-frame Seedance 2 modes. Public URLs and `asset://` IDs are accepted.
`reference_image_urls`, `reference_video_urls`, `reference_audio_urls`	array[string]	Model-specific	Seedance 2 uses multimodal reference arrays. Audio-driven routes such as `video_lip_sync` and `omnihuman_1_5` use `reference_audio_urls` for the target speech or voiceover. Public URLs and `asset://` IDs are accepted where the selected model supports them.
`character_orientation`	string	Model-specific	Motion Control only. Use `video` to follow the motion clip orientation or `image` to keep the uploaded character framing, with durations from 3 to 15 seconds.
`sound`	boolean	No	Only used on models with audio support
`return_last_frame`, `web_search`	boolean	Seedance 2 family	`return_last_frame` requests a separate final image on a best-effort basis; a successful video may not include one. `web_search` enables model-side web search.
`output_format`	string	Seedance 2.5 only	Use `mp4` or `mov`. The default is `mp4`.
`prompt_directory_optin`	boolean	No	Directory discount applies only to selected model families

For real-person footage in Seedance 2, use the UlazAI Seedance 2 asset guide and pass the returned asset:// ID. Direct raw files may fail.

Seedance 2.5 accepts 4-30 second output in 480p or 720p. Multimodal mode supports up to 30 images, 10 videos, and 10 audio files. When a video reference is present, credits are calculated from its measured input duration plus the requested output duration.

Model selector

Choose a model to inspect capabilities and auto-generate request snippets.

Seedance 2.5 model_slug=seedance_2_5

4-30 second generation in 480p/720p with text, first-frame, first+last-frame, or multimodal mode. Supports up to 30 image, 10 video, and 10 audio references plus MP4/MOV output.

Engine: Seedance 2.5

Aspect ratios: 1:1, 4:3, 3:4, 16:9, 9:16, 21:9, adaptive

Durations: 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30 seconds

Quality modes: 480p, 720p, 1080p

Inputs: text , image , video

Credits: 480p 4 no video input: 112, 480p 5 no video input: 140, 480p 6 no video input: 168, 480p 7 no video input: 196, +77 more

Live request snippets

Snippets below follow the selected model preset.

cURL (API key)

JavaScript (fetch)

Python (requests)

Status polling snippets

Poll status by job_id until completed or failed.

cURL

JavaScript (fetch)

Python (requests)

Response examples

Success

{
  "success": true,
  "job_ids": ["8990f846-f6f0-4ff1-a298-d0a72d49e7ad"],
  "credits_remaining_total": 438
}

Validation error

{
  "success": false,
  "error": "Invalid duration for this model"
}

Insufficient credits

{
  "success": false,
  "error": "Insufficient credits",
  "credits_required": 220,
  "credits_available": 94
}

API-key video endpoint (legacy)

If you need API-key based video calls, use /api/v1/generate/video/. That path supports a limited model set and is documented under legacy video mapping.

Open legacy video mapping Open status workflows Open video model matrix

Build one video API flow that survives queued jobs

A video request is asynchronous: submit the prompt and media, save the job ID, then poll until completion or failure. This keeps long renders out of the web request and makes retries traceable.

Validate file type and size before upload, expose the chosen model to the user and record the final error. Never retry a paid job automatically without checking whether the first request already started.