ComfyUI and Flux 2: beginner setup on Windows

Jonathan Winbush walks through ComfyUI setup on a freshly formatted Windows laptop, then uses a Flux 2 template for image editing. He downloads the Windows portable package through GitHub, extracts it and launches run_nvidia_gpu. The interface runs in a browser while generation uses his computer. He notes that CPU execution is possible but describes it as extremely slow.

The setup includes Git and the legacy ComfyUI Manager rather than its version four mode, which Winbush says he found harder to install. He demonstrates updating through Manager and restarting ComfyUI. His preference for the portable package over the desktop application reflects problems he reports encountering himself.

For local AI generation, he filters templates by Runs on ComfyUI, then downloads a text encoder, VAE and diffusion model into their respective folders. The demonstrated downloads include an approximately 8 GB text encoder and a 521 MB VAE. He also explains the Hugging Face account and access agreement required for the gated model shown.

The editing example combines a coffee-can photo with a portrait to replace the can's logo. Repeated runs produce different results, and the first attempt loses details. A later single-image prompt turns the can around but produces unreadable text on its back. The tutorial closes with template filters and a reminder to watch model storage use.