comfyui-krea2edit

Why we included this project

ComfyUI users who want Krea 2 to edit images from a text instruction, instead of just generating new ones, will find this node pack does exactly that. It wires the source image into the model twice, as VAE latent tokens for appearance and through the Qwen3-VL text encoder for semantic grounding, which is how the Krea 2 Identity Edit LoRA was trained. That dual path is what lets prompts like "recolor the car to matte black" or "the man on the left" actually stick to the reference. The README documents the two nodes with a minimal wiring example and a ready workflow file, and it is honest about the practical details: aspect-ratio handling, Turbo versus Raw step counts, and VRAM behavior. The training code is public too, which helps if you want to understand or retrain the LoRA. This is a focused tool for people already at home in ComfyUI's node graph, not a turnkey app.