Comfy Portal

AI Assistant, rebuilt • Replies now stream in as they're written, and you can stop one mid-answer • Every change the assistant makes to your workflow is shown as it happens • Works with 17 providers — OpenAI, Anthropic, Gemini, DeepSeek, Groq, xAI, Mistral, Cohere and more — plus any OpenAI-compatible endpoint, so local setups like Ollama and LM Studio work too • Fixed long prompts being destroyed. Prompts over 120 characters were cut short before the assistant saw them, so its edits overwrote the rest • Custom instructions from Settings are now actually applied New history gallery • Full-screen gallery with multi-select to save, share or delete • Swipe left and right to move between items in the preview • Fixed videos playing with sound over a black screen • Fixed tapping play on a video closing the preview instead Nodes and models • Nodes without a built-in editor now render from your server's own definitions instead of showing nothing • Fixed model pickers showing nothing selected for models that were set and working. Options now come from the server, so custom node folders work too • Load Image can reuse an image already on your server instead of uploading it again • Added support for ComfyUI-MiniMaxH3-Easy • Photos are converted to JPEG before upload, fixing HEIC images being rejected by the server • Fixed very large seed values losing precision Reliability • Generations keep running and reporting when you leave the run page or background the app, and results are recovered when you come back • Fixed the fixed-seed setting never reaching the workflow • Fixed progress occasionally sticking at 100%