Comfy, the developer of ‘
Comfy Router is live
One API for frontier image, video, 3D, and audio models.
Same model string. Same arguments. No new SDK, no new key, no redeploy.
What Comfy Router Gives You:
→ Explicit routing. You name the provider, we call that provider. It's down? The request fails… pic.twitter.com/8ZBw7ljUdP
Comfy Router: One API for Frontier Image & Video Models – YouTube
In the case of closed models, APIs are provided by a single provider, such as Anthropic for Claude or Google for Gemini. On the other hand, in the case of open models, where models can be freely downloaded, APIs for the same model are often provided by multiple providers, and there are differences in price and communication quality from provider to provider. In this situation, OpenRouter, an AI gateway service, is gaining support from users, as it allows users to manage multiple AI models and providers on a single platform.
Comfy Router by Comfy can be described as the ‘image, video, audio, and 3D generation AI version’ of OpenRouter. When integrating content generation AI into an app or service, it is necessary to prepare APIs, API keys, SDKs, etc. for each model and provider used, but Comfy Router allows you to consolidate access to models into a single API.
Comfy Router runs on the Comfy developer platform, allowing you to access multiple cutting-edge models from a single location. A key feature is the ability to switch providers when the same model is offered by multiple providers. If access is restricted, you want to prioritize a cheaper option, or a particular provider is temporarily busy, you can easily switch to another provider and continue generation by simply changing a small part of the API key.

Comfy explains the following three main uses for the Comfy Router:
Product features that utilize a single model
For example, if you're running an image generation function using the developer-oriented media generation platform ‘ fal ‘ on your own service, and access suddenly surges, causing you to reach fal's rate limit, processing would normally be restricted by the rate limit. However, with Comfy Router, if that model is also available from another provider, you can switch the same API call to that provider by simply changing one parameter. There's no need to install a new SDK, obtain a new API key, or redeploy your service.
• A pipeline combining proprietary workflows with models from others.
Workflows created with ComfyUI run on the Comfy API, while state-of-the-art video generation models run on ComfyRouter. Both processes can be executed using the same code and API key, and the execution history can be managed in the same place.
– Large batch processing exceeding API usage limits
By adding hundreds of jobs to a queue using ‘submit’ and maintaining their respective IDs, you can retrieve results as they complete. Since you don't need to keep the connection open while waiting for processing, you can avoid methods that involve waiting for a large number of jobs to finish with retries.
Initial access includes ByteDance's video generation AI ‘ Seedance 2.5 ‘, MiniMax's video generation AI ‘ MiniMax H3 ‘, Google's image generation AI ‘ Nano Banana Pro ‘, OpenAI's ‘ ChatGPT Images 2.0 ‘, Black Forest Labs' ‘ Flux ‘, and Kuaishou's video and image generation AI ‘ Kling ‘. More AI models are expected to be added in the future.
Comfy stated, ‘ComfyUI provides users with control over every step in the image and video production process. All nodes are visible, and all parameters are adjustable. That's why over 3 million creators use ComfyUI. However, calling models via third-party APIs always meant giving up those advantages. You were sending requests to machines you couldn't control and just accepting whatever came back. Most routing layers further widen this gap by automatically selecting providers and calling it optimization. With Comfy Router, you can take control simply by choosing your models and providers. You can always see which router is running.’
Comfy Router does not require a subscription and uses the same Comfy credits used for partner nodes, running Comfy Cloud workflows, and other supported model calls. Pricing per model can be found in the model catalog . In the future, they plan to release routing strategies that automatically select providers based on factors such as ‘cost-first’ or ‘speed-first,’ as well as features that automatically match workflows based on requests.
Related Posts:
<< Next
The indigenous population of Australia before colonization may have been seven times larger than previously estimated.
Prev >>
A special video has been released announcing the start of the theatrical film project for the anime ‘Welcome to Demon School! Iruma-kun’.
in AI, Video, Posted by log1e_dh


