ComfyUI Full Workshop — first workshop from ComfyAnonymous himself!

Summary

• ComfyAnonymous (creator) and team present ComfyUI, an open-source node-based canvas for generative AI across image, video, audio, 3D and text on Nvidia/AMD/Intel hardware plus closed-source API models.
• Traction: top-150 GitHub repo with 78k stars, 3-4M active users, 20k daily downloads, 22k community custom nodes from 3k devs; adopted by Amazon, Apple, Tencent, Netflix and most visual-gen startups.
• Sharability is the secret: every generated image/video carries embedded metadata so dragging it back into the canvas restores the full workflow — fueled viral growth.
• ComfyUI splits diffusion into explicit nodes (diffusion model, CLIP text encoder, VAE, sampler, decode, save) so users can patch the pipeline with custom nodes — e.g., a CFG trick node that modifies negative-prompt subtraction during sampling.
• Workshop demos basic SD1.5 then SDXL workflows locally; mentions video workflows using Wan 2.1 (~60GB models) as currently the best open video model.