Video generation API — 8 models, one submit/poll flow
Kling 3.0 / 2.6, Seedance 2.0 / 1.5 Pro, Google Veo 3.1, Wan 2.5 / 2.2 and Hailuo H3. Submit a prompt, poll the task, download the clip. The clips below were generated through this API.
Crazyrouter is an AI API gateway that serves 8 text-to-video models — Kling, Seedance, Veo, Wan and Hailuo — through one asynchronous video endpoint with one key, billed per generated clip from $0.02, with real sample clips on every model page.
Real clips from these models
Generated through this API, 2026-09-28Text-to-video models, by price per clip
| Model | Vendor | Crazyrouter price (per generation) | Discount | |
|---|---|---|---|---|
| Wan 2.2 T2V Plus wan2.2-t2v-plus | Alibaba Cloud | $0.02–$0.10 / second | — | Try |
| Kling 2.6 aigc-video-kling-2.6 | Kling AI | $0.0357–$0.268 / second | −15% | Try |
| Wan 2.5 T2V wan2.5-t2v-preview | Alibaba Cloud | $0.0429–$0.143 / second | — | Try |
| Kling 3.0 aigc-video-kling-3.0 | Kling AI | $0.0714–$0.143 / second | −15% | Try |
| Hailuo H3 hailuo-h3 | Hailuo (MiniMax) | $0.0714–$0.251 / second | — | Try |
| Google Veo 3.1 aigc-video-gv-3.1 | $0.20–$0.60 / second | — | Try | |
| Seedance 1.5 Pro doubao-seedance-1-5-pro | ByteDance | $1.14–$2.29 / 1M tokens | — | Try |
| Seedance 2.0 doubao-seedance-2-0 | ByteDance | $4.00–$6.57 / 1M tokens | — | Try |
Prices come from the live billing table and already include the Crazyrouter discount; list prices are each vendor's public rate, for comparison only.
What is a text-to-video API?
A video generation API takes a prompt (and optionally a reference frame) and returns a short clip, usually 5–10 seconds, after an asynchronous job. Resolution, duration, motion quality and whether audio is generated are what separate the models; price is per clip or per second.
How to choose
Kling 3.0, Veo 3.1 (with audio), Seedance 2.0. Best motion coherence and camera control.
Wan 2.2 T2V Plus, Hailuo H3, Seedance 1.5 Pro. Iterate on prompts cheaply before a final render.
Kling 2.6 and Seedance for turntable-style shots from a product frame (see image-to-video).
Veo 3.1 generates synchronized sound; the others return silent video.
Generate a clip
Swap model for any row in the table; the example uses wan2.2-t2v-plus.
curl https://api.crazyrouter.com/v1/video/generations \
-H "Authorization: Bearer $CRAZYROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"wan2.2-t2v-plus","prompt":"aerial shot of a coastline at sunrise, slow push-in","duration":5}'
# then poll: curl https://api.crazyrouter.com/v1/video/generations/$TASK_ID -H "Authorization: Bearer $CRAZYROUTER_API_KEY"FAQ
How does the video endpoint work?
POST https://api.crazyrouter.com/v1/video/generations returns a task id; GET /v1/video/generations/{id} until status is succeeded, then download the URL.
How long does a clip take?
Typically 1–4 minutes depending on model and resolution; the sample cards show measured generation time.
How is video billed?
Per generated clip (or per second at the resolution-specific rate); failed tasks are not billed.
Which video model is cheapest?
Wan 2.2 T2V Plus at $0.02 per clip as of the latest sync.
Can I try before paying?
Video runs need an account because each clip costs more than an image; signup adds $2 of Playground credit.
