AI video generation API
Call video models from several providers with one API key. Video requests are asynchronous: submitting returns a task id that you poll for the result. Every model page has a verified request example, parameter ranges and response fields.
How it works
- Sign up and create an API key in the console; send it as Authorization: Bearer.
- Submit a task to POST /v1/videos or /v1/video/generations with the model ID, prompt, duration and references.
- Poll every 3–5 seconds and save the video as soon as it succeeds; result links expire.
Available models
- animate-motion-transfer — Animate Motion Transfer transfers movement from one reference video to a single character image and produces a video. It is not text-to-video and needs no prompt.
- Kling 4.0 Flash — Kling 4.0 Flash supports text-to-video, image-to-video, and multi-image references for fast 3–20 second 720p video generation.
- Motion Video — Creative model: deepseek/deepseek-v4.1-flash. Turn a prompt into a 15-60 second motion-graphics video with animated typography, shapes and transitions. Charged per completed video based on actual creative work, duration and resolution. Historical estimates appear when enough reliable samples are available; failed generations are refunded.
- Seedance 2.0 — Seedance 2.0 has 18 public models: Standard, Fast and Mini tiers, each with text-to-video (t2v), first/last-frame image-to-video (i2v) and mixed image, video and audio references (multi). IDs without global run in China; IDs with global- are the international versions.
- Seedance 2.5 Standard — Seedance 2.5 Standard supports text-to-video, first/last-frame image-to-video and mixed image, video and audio references. Use the exact t2v, i2v or multi model ID for the chosen workflow.
- Wan 3.0 — Wan 3.0 offers text-to-video (t2v), first/last-frame image-to-video (i2v) and multi-reference generation (r2v). wan-3.0 runs in China, prime is the faster tier and global is the international version.
FAQ
Is it OpenAI-compatible?
Authentication and the base URL follow OpenAI conventions. Video generation is an async task you submit and then poll; each model page lists its exact fields.
Can I see the price before submitting?
Yes. POST /v1/video/generations/estimate returns an estimate, and the finished task reports the actual charge.
Am I charged for failed tasks?
No. Failed generations are refunded, and the error is returned in the task result.