Wiggly fingers

[Ashdevz]

Those wiggly fingers need to be optimised more by the neural network to make them more coherent, even with negative prompts such as (bad-hands-5), (badhandv4), (deformed hands), and (extra fingers), are still producing below-par results.

Will there be any fixes in the future? Any current development in that area, perhaps a similar neural net like Restore Faces could be created for hands?

Would love to hear you thoughts...

Thanks

[AlluraHikaru]

controlNet depth exists and with https://github.com/jexom/sd-webui-depth-lib it's easy'ish to generate hands, or fix hands. Theoretically, just like face detection it's possible to detect and fix hands with controlnet depth automatically, but i would rather just select hand type manually or create hand type for depth from scratch to fit the picture better.

[Ashdevz]

Hopefully control net will give me the results I'm looking for, I'm downloading the extensions for it. Thanks for your reply, appreciate the feedback and it has put me back on the 'path' ha

Cheers

[Ashdevz]

Don't know if you know about embeddings, I've added them and that seems to have done the trick!

[arvindbr95]

Hey @Ashdevz, can you share more about using embeddings to fix hands? I'm facing this issue too and am looking for a solution.

Thanks!

[Ashdevz]

You can get the embedding for bad hands here https://huggingface.co/yesyeahvh/bad-hands-5/tree/main and put this file in your embeddings folder and then use it in negative prompts like this (without the spaces) < bad-hands-5 > can also be weighted < bad-hands-5:0.8 >, < bad-hands-5:1.0 >, < bad-hands-5:1.2 >, etc. Be mindful that embeddings can change the style of the image. ControlNet is a much better solution if you have the GPU power 8gb+.

Hope this helps

[arvindbr95]

Thank you!