Try the model before you write the client
Chat, generate an image, render a video — in the browser, against the same gateway and the same deployment your code will call. If the request works here, it works there.
A real request, with the training wheels off
- 01
Six modes, chosen for you
Chat, Completion, Image, Image Edit, Video and Image-to-Video. The mode follows the model's declared input modalities, so there is no wrong one to pick.
- 02
No key juggling
Entering issues a temporary session key that lives 30 minutes and renews itself. Your real workspace keys are untouched — and session keys are deliberately useless in a script.
- 03
Real constraints, not a demo
Size and resolution pickers only offer what the model actually supports, validated exactly as the API validates it. A request refused here would have been refused there.
- 04
It leaves with the code
Every panel generates the matching curl or Python snippet for your own endpoint. The step from trying something to shipping it is a copy.
What each mode actually calls
The same OpenAI-compatible surface your client will use.
| Mode | Endpoint | Path |
|---|---|---|
| Chat | /v1/chat/completions | Sync, streaming supported |
| Completion | /v1/completions | Sync |
| Image, Image Edit | /v1/images/generations, /v1/images/edits | Sync within 30s, or async |
| Video, Image-to-Video | /v1/videos, /v1/videos/generations | Always async |
Pick a card and start the pod
Sign up, choose the GPU, and the pod is yours in under two minutes. It bills by the hour and stops when you stop it.
Already a user? Invite friends and earn 15% of their first top-up.