Control image generation by edge maps generated with Edge Drawing Parameter Free (edpf). This is similar to ControlNet Canny but uses a modern algorithm for edge-detection which requires no parameter tuning (Canny was invented in 1986, Edge Drawing in 2012). You can use the gradio demo or edpf.py script to generate edge maps until a pre-processor is implemented in your favorite Stable Diffusion UI.
Feedback is welcome! I'm still improving this model and you can help me by generating simple usecases and discuss the results.
The case for a new edge model
Do you find all these settings for canny confusing and time-consuming?
Do your default edge maps have noise, artefacts and missing edges?
All your images come out a mess like they were drawn by a human?
Well look no further!
With Edge Drawing Parameter Free you can create a masterpiece with a single mouse click!
No tuning! No mistakes! No frustration!
By combining recent advances in deep learning, cloud computing and block chain technology(?) we created the EDPF control net model just for you.
And the best part: it's free! Parameter Free!
If you want to train your own control net see my article Play in Control - ControlNet training setup guide!
Description
FAQ
Comments (15)
This looks promising. Canny is too complicated for me
you can also use the edpf script to generate edge maps for the canny model. for result comparison look here: https://github.com/lllyasviel/ControlNet/files/12700786/control-edgedrawing-cv480edpf-drop0-fp16-checkpoint-45000.zip and compare the images "canny-canny" with "canny-edpf".
Sir and/or Madam or.. whatever.. ,
Your clear bias has cast great doubt on your integrity as a __self_identifier_3__ . The fact that you choose to segregate a portion of the populous, and thus depriving those passionate few who's only goal is to find comfort in the solace of a <Formula 1 racing cars> waifu, who can only hang their heads in shame... self.loathing ... no no, what was that other thing.. oh yeah! ... disappointment!
I am NOT a robot! I am a NaN!
馃枛馃ぃ馃憤
apparently some people do read ;)
@GeroldMeisinger聽three ghosts of formula 1 history will visit you, the first one being the ghost of f1 past: bernie ecclestone! repent as long as you can!
Lovely work! I agree that it produces better result than Canny.
Although you can't beat the swiftness of having the canny presets.
Any chance for a WEBUI preprocessor?
I would love to use this as an option <3
Great work here, and thanks for writing the amazing article about ControlNet training!
One day...
Thank you!
> webui preprocessor
I will look into it once it's mature enough of course but I think a Gradio demo will be quicker in the meantime. I don't know what Mikubill's stance is on exotic control nets and how long it will take to get it through the pipeline. You could investigate into this if you want to help, open a feature request or even better provide a pull request (if you can code). Some pointers on how to implement here and here .
gradio demo on hf spaces https://huggingface.co/spaces/GeroldMeisinger/edpf
@mnemic new version!
@GeroldMeisinger聽Hype! I'll give it a spin in a week or something!
Have you tried lineart-realistic controlnet model comparison? thanks.
I don't know about lineart-realistic yet, have to look into it. But I would love to see comparisons and evaluations!
@GeroldMeisinger聽please refer to the guide. lineart realistic is among others of controlnet 1.1
I see, I was confused by the "-realistic" suffix and assumed it was a custom model. lineart was trained on https://huggingface.co/spaces/awacke1/Image-to-Line-Drawings and fills a different use case: "drawing from scratch" whereas the edge detectors are meant to be "derived from existing images".
@GeroldMeisinger聽sorry for mixing you up, I just tried to ask if you tested your model in the comparison with lineart realistic or not. got your point about different usage, thanks.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.