Yes, I called it lazy and i am proud of it! Why?
Because all you need to do is literally 3 things.
HOW IT WORKS
1 add the image you want to inpaint in group 0
2 jump to group 3 and in prompt input type the image segment you want to mask (in my example I use dress)
3 write a short prompt regarding your inpaint expectations in group 4, in my example i wanted to write "#bonbonero" on her dress to promote my Twitter format
Description
FAQ
Comments (18)
Hey, love the idea and layout for the workflow.
As someone completely new and learning Comfy for the first time, how do I get the image into the flow that I want to inpaint? In doing so will that remove all the issues with the 'red nodes' or do I need other files and models above the standard model, clip and vae?
Thank you for your help
Hi, no. If you see "red nodes", this means you need to install the missing custom nodes via manager.
Open it and click on install missing custom nodes, then restart and refresh.
Before diving into an inpaint workflow, I'd suggest you to learn the basics. it will help you understanding the fundamentals, then after you can learn something more complex.
This is my humble suggestion.
@KayKayTrax Thank you very much for your reply, I appreciate the assistance. You are right of course, getting carried away with all the 'shiny' options and potentially skipping useful fundamentals. Will look to build up knowledge from the beginning first ;)
Thank you! This is an excellent workflow! Works like a charm and love the way you've laid out/organized the nodes 🤌.
I have a question about expanding upon this...
I'm trying to integrate a flux depth controlnet into this workflow to help follow the entire body more consistently during inpainting, while only inpainting on the torso. I had this working in an SDXL workflow and was getting 90% accuracy/consistency with inpainting. Haven't had such luck with trying to adapt this methodology to Flux yet. Any ideas on how this could be done?
I would try with Instant-X. You can connect it to the positive conditioning. Hope that helps :D
@KayKayTrax I'll give it a go, thank you!
Hi to you too! :P
the error means you ran out of memory.
You can try to use a GGUF model (Q8 or Q6 or lower), use them with their loader
https://github.com/city96/ComfyUI-GGUF
GGUF models can be found here:
flux1-dev GGUF
flux1-schnell GGUF
@KayKayTrax Thx, will try!
@zerocool22 update. I had a similar problem too.
1 check the ref image is not too big
2 check if the onnxruntime gives you problems. You might need to uninstall the misatching version and reinstall the previous one
pip uninstall onnxruntime
pip uninstall onnxruntime-gpu
pip install onnxruntime==1.18.0
pip install onnxruntime-gpu==1.18.0
Thanks for sharing. Well, it looked good but... there's always something breaking, right :/
GroundingDinoSAMSegment doesn't work :
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument mat2 in method wrapper_CUDA_mm)
So for now I just paint a mask manually, but let's be honest this all doesn't work very well, I get better inpainting results in Forge with less trouble...
Thanks for sharing your experience! It sounds like the error you're encountering is related to your device setup—when tensors are on different devices (like CPU and GPU), that can cause issues. Since the workflow is functioning for others, it’s likely a configuration issue rather than a problem with the workflow itself. If you’d like, I can help you troubleshoot to ensure everything is running smoothly on your end. Comfy can be a bit tricky, but once things are set up correctly, it should work well!
@KayKayTrax You don't really need my autorisation to help me, if you can help please do, thank you x) Of course I'd like help that's why I commented in the first place lol thinking you might know the issue. With AI tools I now spend my whole life fighting Python errors it's painful I'm no developer xD
@Tetsuoo I am not a python dev either but with more context I might help. let me know :D
@KayKayTrax ok... Forget it man... forget it.
hello;
thks for sharing, its a cool and simple workflow.
i try to arange the form and the color of a pillow.
it render it more comfy as i want but the color dosen't change.....did i something wrong?
try to write in the prompt
(blue pillow)::8
or whatever color you need. :)
ok thx; i will try your synthax