Skip to main content
Use GET /v1/models to see which models your team has been granted access to. The full catalog below shows every model the platform supports — contact your administrator to enable specific ones for your token. Video models → POST /v1/videos (async, poll for result)
Image generation models → POST /v1/images/generations (text-to-image, sync) · POST /v1/images/edits (image-to-image, sync)
LLM (chat) → POST /v1/chat/completions (OpenAI-compatible)
STT, OCR, caption → POST /v1/generate (synchronous) Duration ranges are verified against the API as of 17/08/2026.

​
OpenAI Image

Generate images synchronously via POST /v1/images/generations (text-to-image) or POST /v1/images/edits (image-to-image, pass image + prompt).
Run GET /v1/models for the live list — the catalog is updated regularly as new models become available.

​
Chat / LLM (POST /v1/chat/completions)

Standard LLM models use POST /v1/chat/completions with OpenAI-compatible messages[] format — the same interface as the OpenAI API. Check capabilities.endpoint from GET /v1/models to confirm which endpoint a model uses.
Run GET /v1/models for the full live list — the catalog is updated regularly as new models become available.

​
Generate — STT · Vision · Caption (POST /v1/generate)

Non-LLM models use POST /v1/generate with a prompt (or audio/image/video) field and return the result immediately.
Run GET /v1/models for the full live list — this table shows the most common models.

​
Kling

​
Kling v3.0

Supports 3s or 15s only (no other value accepted). Supports cfg_scale, sound, shot_type, end_image, multi_prompt.

​
Kling v3 Turbo

Supports 5s, 10s only. Faster and cheaper than v3.0.

​
Kling Video O3

Supports 3s or 15s only (no other value accepted). Supports cfg_scale, sound, shot_type, end_image, multi_prompt, images (reference-to-video).

​
Kling Video O1

Supports 5s, 10s only.

​
Kling v2.6

Supports 5s, 10s only. Supports sound, cfg_scale.

​
Kling v2.5 Turbo

Supports 5s, 10s only. Supports last_image (end frame).

​
Kling Lipsync

​
Seedance

​
Seedance 2.5

Supports 4–30 seconds (any integer), or -1 for auto-length. Supports generate_audio, resolution (720p, 1080p, 4k), end_image, images[].

​
Seedance 2.0

Supports 5–10 seconds (any integer).

​
Seedance 2.0 Mini

Supports 5–10 seconds (any integer).

​
Seedance 2.0 Fast

Supports 5–10 seconds (any integer).

​
Seedance v1.5 Pro

Supports 5–10 seconds (any integer).

​
Seedance v1 Pro Fast

Supports 5–10 seconds (any integer).

​
Veo 3.1

Fixed 8s clips, 16:9 only.
duration and aspect_ratio are not honored by this model — every request returns an 8s, 16:9 clip regardless of what you pass. resolution (720p default, 1080p available) does work. Verified against the live API on 21/09/2026.

​
Omni

Supports 3–10 seconds. duration, aspect_ratio (16:9/9:16), and resolution (720p default, 1080p available) all work as documented — verified against the live API on 21/09/2026.

​
Wan 3.0

Supports 2–30 seconds (any integer), or -1 for smart/auto-length. duration, aspect_ratio, and resolution (up to 4k, ESR upscale tiers included) all verified against the live API on 25/09/2026.

​
MiniMax H3

Supports 4–15 seconds (any integer). duration, aspect_ratio, and resolution (480p/768p/2K, plus ESR upscale tiers) verified against the live API on 25/09/2026.
If aspect_ratio is omitted, 16:9 is sent explicitly rather than left unset — this model rejects its own default for text-only requests.