Autonomous AI agent in the ComfyUI sidebar - local-first, bring any LLM. Run it on your Claude or ChatGPT subscription (no API keys, no extra LLM costs), on a Kimi K3 / GLM / OpenRouter key, or fully offline on free local models via Ollama, LM Studio, or llama.cpp. The agent drives your live canvas at full parity: add, wire, move and tweak nodes with Ctrl+Z undo, load whole workflows and installer packs in one shot, generate images, video and audio, browse CivitAI, and run your workflow - asking before it spends paid API credits. The sidebar UI for comfyui-mcp, the local-first, agent-native control plane for ComfyUI. The panel speaks 12 languages, matching your ComfyUI language automatically: English, 한국어, 日本語, 中文 (简体), 中文 (繁體), Русский, Français, Español, Português (BR), Türkçe, العربية, فارسی.
https://github.com/artokun/comfyui-mcp-panel
git clone https://github.com/artokun/comfyui-mcp-panel
comfy node install comfyui-agent-panel
为 DIT 模型设计的简单逻辑神奇放大节点工作流。通用适用于 Flux、Hunyuan、SD3：把初始图像简单平铺成碎片，再用 image-interrogator 为每图块获取提…
用 VACE 控制潜空间把短视频或单张图像扩展为更长、时序一致的片段。运行多个扩展循环；每循环可有自己的提示词与可选 LoRA、lastframe 支持：逐步演变场景而无需重建图。