Appearance
DeepInfra
DeepInfra (/media/deepinfra) is a cost-optimized inference hub for open generative models. The studio covers image, video, and text-to-speech generation through DeepInfra's native per-model inference endpoint.
Where to configure
Configure your DeepInfra AI provider once under Settings → AI. DeepInfra is a universal-credential provider: the same API key drives both LLM and media generation, so no separate Settings → Media entry is needed. See Settings for provider setup.
Tabs / operations
| Tab | Operation | Key fields | Produces |
|---|---|---|---|
| Text → Image | Image | Model (FLUX.1 schnell/dev), prompt, width, height, steps, seed | Image(s) in /files |
| Text → Video | Video | Model (Veo 3.1 / PixVerse V6), prompt | MP4 clip in /files |
| Text → Speech | Audio | Model (Kokoro 82M), text, voice preset | Audio file in /files |
Generation flow
All three tabs post to DeepInfra's synchronous /v1/inference/{model} endpoint. The artifact returns inline, is saved to the Media Library, and appears in the Render Queue with Edit in Designer and Post hand-offs. See Media Studios overview for the shared flow.
Caveats
- Because DeepInfra has no clean per-modality model catalog, the model dropdown uses curated lists plus a free-entry field; type any DeepInfra model path if yours is not listed.
- The adapter probes common response keys (
images,image,audio,video_url,output) to extract the artifact. Unusual model response shapes may need adjustment against a live key. - All operations complete synchronously; there is no webhook or background poll path.
Related docs
- Media Studios overview — shared render-queue and hand-off flow.
- Media Library — where finished images, video, and audio are saved.
- Settings — configuring AI providers.
Verified against v1.0.0