Skip to main content
Comfy Router serves every model on one route, POST /v2/models/{provider}/{model}. A few of those models can be served by more than one provider, and the model_provider query parameter picks which one runs the call. The model ID, the request body and the response shape do not change. See model_provider, strict_mode and fallback_provider in the API reference.

​
Comfy (direct)

Every model in the model catalog is served by Comfy Router directly. This is what a call with no model_provider gets, and it is the only route for every model not listed below.

​
fal

  • Dreamina Seedance 2.0 260128: byteplus/dreamina-seedance-2-0-260128, served as fal/fal-seedance-2.0 with ?model_provider=fal
  • Dreamina Seedance 2.5 260628: byteplus/dreamina-seedance-2-5-260628, served as fal/fal-seedance-2.5 with ?model_provider=fal
  • Nano Banana 2: vertexai/gemini-3.1-flash-image, served as fal/fal-nano-banana-2 with ?model_provider=fal
  • Nano Banana Pro: vertexai/gemini-3-pro-image, served as fal/fal-nano-banana-pro with ?model_provider=fal
  • GPT Image 2: openai/gpt-image-2, served as fal/fal-gpt-image-2 with ?model_provider=fal

​
WaveSpeed

  • Nano Banana Pro: vertexai/gemini-3-pro-image, served as wavespeed/wavespeed-nano-banana-pro with ?model_provider=wavespeed
strict_mode defaults to false, so a request written against the native model’s schema is translated into the alternate provider’s own schema, and the response is translated back. Any native field that cannot be expressed on that provider is dropped and named in the X-Comfy-Router-Dropped-Params response header.