Here's an inpainting workflow for Flux that actually works without eating all your VRAM or messing up the original image colors (kinda)
Runtime profile
Source description
Here's an inpainting workflow for Flux that actually works without eating all your VRAM or messing up the original image colors (kinda)
Instead of processing your entire image, it isolates the exact area you want to change, inpaints it at high resolution, and stitches it back together. You can either draw a mask manually, or just type what you want to mask using the built-in SAM2 node.
Requires: Crystools, KJNodes, InpaintEasy, ComfyUI_essentials, rgthree, and comfyui-rmbg (for SAM2).
AI-generated commentary
AI-generated explanation based on source and configuration details. Suggestions are clearly labeled.
Ultra Inpaint — v2.0 is a Flux image-inpainting workflow for changing a selected area of an existing image, using either a manually drawn mask or a text-described area, and compositing the edited area back into the original image.
It isolates the selected area, inpaints it at high resolution, and stitches it back into the original image.
You can draw the mask manually or describe what to mask through the built-in SAM2 node.
Input an existing image and receive an edited image with the selected area composited back into it.
The workflow is based on Flux.2 Klein 9B and published by SneakSexyGirl.
The listed version is v2.0, classified as Workflows.
The workflow lists these node packs: comfyui-inpainteasy, comfyui-easy-use, comfyui-workflow-encrypt, ComfyUI-Crystools, rgthree-comfy, comfyui-kjnodes, ComfyUI-GGUF, comfyui_essentials, and comfyui-rmbg.
The listed model files are EVA02_CLIP_L_336_psz14_s6B.pt, flux-2-klein-9b-Q8_0.gguf, flux2-vae.safetensors, fluxKleinFP8_flux2Klein9bFp8.safetensors, and Qwen3-8B-Q5_K_M.gguf.
A desaturation workaround is included before compositing to address possible brightness or saturation changes in the inpainted area.
The approximate desaturation range is 0.05–0.20 and varies by image.
Before use, verify that your setup includes these exact model files: EVA02_CLIP_L_336_psz14_s6B.pt, flux-2-klein-9b-Q8_0.gguf, flux2-vae.safetensors, fluxKleinFP8_flux2Klein9bFp8.safetensors, and Qwen3-8B-Q5_K_M.gguf.
Suggestion · not verified
Before use, verify which ComfyUI-GGUF package your setup should use.
Suggestion · not verified
Does this need to be edited?
Sign in to send an edit request.
Sources
1 sourceSource excerpts
2 excerptsSource context: 1064 downloads · Type Workflows · Base model Flux.2 Klein 9B
Estimated VRAM requirement
Estimate unavailable
160 MB across 1 of 5 model files. Model file total + 25% loading overhead + 2 GB execution buffer, rounded up.
Requirements
14 requirementsText encoder · Hugging Face · Arctic1998/EVA02_CLIP_L_336_psz14_s6B
flux-2-klein-9b-Q8_0.gguf
Not resolvedCheckpoint · Unknown
VAE · 160 MB · SAFETENSORS · Hugging Face · darkmaniac7/tokforge-bonsai-image-4b-gguf
fluxKleinFP8_flux2Klein9bFp8.safetensors
Not resolvedCheckpoint · Unknown
Text encoder · Hugging Face · Mukilan06/Qwen3-8B-Q5_K_M · Qwen3 8B Q5 K M
Node pack · Registry
Node pack · Registry
Source context: 981 downloads · Type Workflows · Base model Flux.2 Klein 9B
Node pack · Registry
ComfyUI-GGUF
PossibleNode pack · Registry
Node pack · Registry
Node pack · Registry
Node pack · Registry
Node pack · Registry
Node pack · Registry