Inpaint workflow V2
GUIDE FOR V2 WORKFLOW IS LOCATED HERE
06/12/2025
-Added ability to control mask opacity.
-Changed sampler to prevent problems with generation.
-Adjusted guide.
04/12/2025
-Added ability to do colormatch. Changing colormatch setting to 1 will try to match the color of the image.This is good if using VAE that changes color between encode/decode cycle or using different lora/model.
Older:
-Added differential diffusion
-Cleaned the workflow
-Created guide which can be found HERE
-This should work with 1.5 too, did not test.
Inpaint workflow XL V1.1
Due to request updated to work with XL. It is somewhat barebones compared to 1.5. Does not have segmentation or controlnet models currently.
I have a bit outdated comfyui, let me know if it is throwing some errors.
Description
FAQ
Comments (50)
Thanks again for your amazing workflows! 👍
(～￣▽￣)～ Thanks. Feedback welcome if you find something problematic
I tried this workflow and it works great
Glad to know (～￣▽￣)～ Let me know if you have some ideas for improvements
I like your workflow. I am inpaint anime characters and I want to know is there a way to make the skin color more align to the original one.
I am using 2.02, I can't get 3.1 to work, it says ScheduleToCond is missing.
@goldenluluz Hey, test if the new XL one works with 1.5.
@PotatCat Hello, I am using the XL V2 version with NoobAI now, and it is perfect, thank you for working!
@goldenluluz (～￣▽￣)～
I heard that if you use ip-adapter, you can transfer the style from the original picture to the generated picture. Have you considered adding ip-adapter controlnet into your workflow?
@goldenluluz It's been a while since i played around with ip-adapter, gotta see if i can find time and interest to do that. Tho i am not sure if there is much use for it in inpainting. 🤔
Hey I was trying to remove some jewellery from an an anime character, is there any workaround or like any specific prompt to remove something or if I need to try something else
Thank you
Hey, You should describe in the prompt what you want in the selected part of the image. If it's small part of image, you could try to edit away the jewellery with photo editor like photoshop or krita. Also could try to inpaint it after you have edited the jewellery away, to fix any inconsistency etc..
@PotatCat Thank you
hello im generating these black splodges on the masked area i dont know what im doing wrong
Hey, did you check the guide in https://civitai.com/articles/9971 ?
@PotatCat thanks working
Currently a VAE is required, but I never use VAE in any of my generations.
How can I disable VAE?
If you want to use the baked in vae from model, you can drag the line from model VAE input into where the one from the vae loader is going. With comfyui you need to have some sort of VAE input.
@PotatCat Follow-up question because I am not used to ComfyUI.
Does ComfyUI have a different way of handling models and prompts?
Because the prompt adherence is awful, it's not doing what I want at all.
@ThatNewDood Yes the weighting and normalization works differently on comfyui compared to forge/a1111 for example. Using high weights not really good idea on comfyui.
Works great!
glad to know (～￣▽￣)～
I have an artwork (a painting). Now I want to create a mockup of living room showing my painting in the frame on the wall.
I will load the living room image with the empty frame on the wall in the workflow, but how I can input my artwork into the frame on the wall?
Tbh, best way to do this would be to edit the painting in the image with photoshop, krita or other image editor.
I don't know if this workflow is best for doing that, as it has some problems with editing full images, as it needs to cut square from the image.
Best inpaint workflow I have used. I keep trying some new ones that pop up but this one by far gives me the best results. Thank you for this, appreciate you!
(～￣▽￣)～Thanks, my goal always was to make it as close as possible to replicate the inpainting from webui like A1111/forge. With the idea of cutting only part of the image you want to inpaint and enlarging it to 1024x1024 for better result.
Doesn't work with latest version of ComfyUI. 😔 Missing Node Types CR Apply LoRA Stack. Error Import Failed in Comfyroll Studio node.
Uh oh, always changing something and breaking my workflows ;( You can replace the lora nodes with another. Just follow the line from lora stacker and replace the CR Apply Lora stack with another lora loader. I am using quite old comfyui currently and usually only update when there is something worth to update for.
@Rev34 Mine is quite old at this point. ### ComfyUI Revision: 2946 [9f4b181a] *DETACHED | Released on '2024-12-18'
Pretty sure it works with some newer one.
@PotatCat But I don't know what and how to replace CR Apply LoRA Stack with, I'm not that experienced in this. And the new version of ComfyUI is extremely important to me because of Chroma (a game changer in the field of generations, which can soon displace all other models).
@Chertilo You can replace it with basic "load lora" or any other lora loader. https://i.imgur.com/qlyNhYK.png fast paint paste on what you need to rewire.
@PotatCat Thank you very much. 😊 Now everything works. 😊
@Chertilo (～￣▽￣)～
Works even with the LoRA stack loader, if you adjust the number of "lora_count" argument from "9" to "3".
@TesterBester Can you show a screenshot of how you connected LoRA Stack node after deleting the non-working CR Apply LoRA Stack node?
@Chertilo I did not change any connections at all from the given example. I just set the "lora_count" value to "3" (from 9) and it worked. You can still use the exact same "lora stacker" node.
Finally went around and updated everything. Seems like the maximum lora count has been reduced to 3 on the stacker for some reason.
Created fixed version with another lora loader.
@PotatCat Thank you very much.😊
@TesterBester With the CR Apply LoRA Stack node, your method didn't work in my case, as it still reported an error with it.🤔 https://postimg.cc/gallery/tsnvBLD
However, all this is now absolutely unimportant. 😌
Hey, I have a question, I see some people use controlnet for inpainting, this workflow doesn't use the controlnet inpainter, I don't understand the differences between your method and the controlnet one? This one https://civitai.com/models/1376234?modelVersionId=1555027
Hey. Controlnet inpainting can help keeping the composition and reduce "bad" inpaints. When i created this workflow there was no tile/inpaint controlnets for XL that could help with inpainting. I have gotten several request on to implement controlnet support. I'll put that on my list to do.
The workflow uses differential diffusion that helps with inconsistency thanks to ability to do alpha opacity masks and it seemed to be enough to reduce the inconsistency of inpainting with higher denoise, so i did not implement controlnet into the workflow back then.
Also the inpainting model you linked, i personally noticed some odd color shifting. I recommend the noob tile model over the inpainting model personally. With start 0, end 0.25-0.4 of the controlnet. Depending on how much you want to change the image.
Don't know if i exactly answered your question. Basically controlnet is extra tool that can help with results.
@PotatCat
Yes, you absolutely answered my question! Thanks a lot. I'm fairly new to this, so I have trouble understanding how half the things work, and the differences between all the options I keep finding. Thanks for the explanation and the tips!
I'm also learning how to properly use your regional prompter, so far it's working amazing! I will upload some pictures when I am happy with my results.
@AciesGecko (～￣▽￣)～I'll look forward on what you manage to make. Also i don't know how new you are to generation, but will clarify: When you use controlnet model, it will use both the checkpoint and the controlnet model. The controlnet model in simple terms gives "instructions" to the generation.
Great workflow! Works like a charm on up to date ComfyUI version after adjusting thr "lora_count" value on the LoRA stacker to "3".
The NoobAI model you suggested in your example did not work that well for me, but (e.g.) the "Smooth Mix Noobai / Illustrious V3" model did the magic.
Thanks!
(～￣▽￣)～nice to know it is still holding even with small hiccups. Inpaint can be pain sometimes on specific cases and some models can perform better than others. Usually if the are you are inpainting is too "zoomed in" it can be hard to get what you want. Usually increasing padding can help on those cases.
Thank you for the XL update. I was running into issues with the previous version when I selected a larger mask area the workflow wouldn't work. I I'm not sure what the issue was. It's like certain size mask interferes with the stitch nodes. It would result in the ksampler not processing the mask region.
It cuts 1:1 ratio from the original image, if the height or width is too small for it to be able to cut 1:1 ratio image it will not work correctly. I did not really find practical solution to this. At that point the best case is to just inpaint with the whole image.
@PotatCat thanks for pointing that out. Is there a setting I could adjust in the workflow that would use the whole image of a section isn't working?
@Sexiam Sorry not in this workflow. and implementing it to this workflow could make it confusing. It is quite simple to do basic inpainting workflow, pretty much only need these apart from normal generation stuff. Using some grow mask/blur will help with the lines too. https://i.imgur.com/Z0WsOq2.png
