Skip to content

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

TabOperationKey fieldsProduces
Text → ImageImageModel (FLUX.1 schnell/dev), prompt, width, height, steps, seedImage(s) in /files
Text → VideoVideoModel (Veo 3.1 / PixVerse V6), promptMP4 clip in /files
Text → SpeechAudioModel (Kokoro 82M), text, voice presetAudio 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.

Verified against v1.0.0

The AI-native social media management platform — postmill.ai