ComfyUI-Image-Inpainting
Nodes:VAE Encode Inpaint, VAE Decode Inpaint, ColorCorrection Inpaint, ImagePreprocess Inpaint, ImagePostprocess Inpaint, Load Model Inpaint, Inpainting (using Model)
Nodes (10)
The depth-map crop nobody searches for, and why it matters
A RealESRGAN wrapper with a footgun hiding in its constructor
The unglamorous resize that keeps masks honest
Making the inpainted patch stop looking foreign
The 'only masked' inpainting trick, wrapped in one node
The LaMa runner — fast object removal, with honest limits
The pack starts here — load the LaMa model that does the heavy lifting
The last node in the pipeline — pasting the surgery back together
The decode half of the pair — back to pixels, right-sized
VAE encode for inpainting, without the padding headache
-
cd custom_nodes/
git clone https://github.com/ltdrdata/ComfyUI-Impact-Pack.git
git clone https://github.com/SherryXieYuchen/comfyui_controlnet_aux.git -
download sd inpainting model "sd-v1-5-inpainting.ckpt" and put it in models/checkpoints/stable-diffusion-inpainting/
download vae "vae-ft-mse-840000-ema-pruned.safetensors" and put it in models/vae/
download controlnet depth model "control_v11f1p_sd15_depth.pth" and put it in models/controlnet/
download lama inpainting model "big-lama.pt" and put it in models/inpaint/
download upscale model "RealESRGAN_x2plus.pth" and put it in models/upscale_models/
download depth model "ZoeD_M12_N.pt" and put it in custom_nodes/comfyui_controlnet_aux/ckpts/lllyasviel/Annotators/
download lcm lora model "pytorch_lora_weights.safetensors" and put it in models/loras/