This Adetailer model will segment speech bubbles, text and watermarks commonly found in training data. Trained this so I could eventually automatically clean images in a dataset. Only tested on Comfy, but should work on other webUIs too. This is a WIP, and I have many things in mind on which could be improved:
Known issues:
make sure you don't set minimum confidence too low, or else undesired objects will be segmented
can misidentify watermarks for text, speech bubbles for logos etc. but this should not matter since they are segmented anyway
Some text that is transparent/partially hidden won't be identified
Trained primarily on NSFW images, may not work too well with comics, images with large/strange fonts etc.
Description
Increased dataset size, better annotation
FAQ
Comments (16)
how do i use this on webui, (im using forge)
I dont use forge, but you need to find the directory of the other adetailer models. If it throws an error try updating ultralytics
@septagon i tried on img2img and it detect some text and watermarks but it doesnt clean it off, the text is still there, what prompt should i write for it to understand that i want to remove texts? i have written "remove text" or "clear" but it just wont do anything
@monicalucci I dont really know if forge can completely remove the text, I use a comfy workflow. Its supposed to function as a simple detector, adetailer itself isnt great at removing things. Check the guide, if you want to spend some time setting up comfy etc
@septagon Bro I am new to Comfy UI. Can please share the workflow. I am having trouble setting it up.
Could you give an example that how to add words in speech bubble?Thx.
This is a detection model, it'll only tell you where the speech bubble is. If you want to add text your best bet is Photoshop
@septagon Thanks
any idea on how to make it so this cleans text on A1111?
tried to do so myself, but ended up switching to comfyui. You could try lama for A1111 and see if you can get that to work with adetailer. Otherwise comfy installation and setup is extremely simple
@septagon last time i tried it caused too many issues with loras and extensions so kinda gave up. Will probably try again at some point.
There's a "Detection" model type now you could change this to, so people can find it easier. I'm commenting this on most Yolo models that are still stuck in "Other".
Thanks man, good to see it's finally an option
Does any one know how to use this in Comfy UI. I am new to it. I dont know how to use it. Please help.
Hi, thanks for your model and workflow!
Can you confirm the V2 Download is V2? Because inside the .zip there's V1
Works like a charm for detecting text to inpaint it!