What features do you most want us to add ?

ResearcherXman commented 8 months ago

We are a small team and cannot handle all feature requests at once. Thus, please tell us what is your favorite feature you wanna us add. Comment under this post. Meanwhile, we encourage all kinds of PRs (fixing bugs, adding new features, etc.) from the community, if you want to take on a feature, please open a new issue naming "Support for xxx" and we will assign you. Thanks for your interests, love you all ❤️.

brunneis commented 8 months ago

~~It would be great to support non-human faces.~~ It would be great to add better support for non-human faces. Congrats for the great job!

Edit: it works on animals, but with many pictures I get a "Cannot find any input face image" error:

Traceback (most recent call last):
File "/root/.pyenv/versions/3.11.7/lib/python3.11/site-packages/cog/server/worker.py", line 217, in _predict
result = predict(**payload)
^^^^^^^^^^^^^^^^^^
File "/src/predict.py", line 173, in predict
face_info = sorted(
^^^^^^^
IndexError: list index out of range

zewolf5 commented 8 months ago

Not sure if its already there, but multiple reference images? I could not find a way to select more than one image in the sample code provided. Topic #33 mentioned use of multiple reference images for better results.

Also is it possible to do some "adetailer magic" like upscaling the part of the image that contains the face, apply hires face-swap and then resize down again to the original image to get a detailed face when not covering the whole image?

I am still new to this, so ignore anything i have misunderstood.

fofr commented 8 months ago

I'd love to see implementations of each of the examples in the paper.

tgohblio commented 8 months ago

SD 1.5 support.

Opens up whole new possibilities with LORAs, AnimateDiff (as of today, it's SD 1.5 only since XL is in beta) and community-trained SD1.5 models. Hint: this will drive more Github ⭐️

nosiu commented 8 months ago

Not sure if its already there, but multiple reference images? I could not find a way to select more than one image in the sample code provided. Topic #33 mentioned use of multiple reference images for better results.

I'm no expert but this should work image = pipe( ... image_embeds=(face_emb1 + face_emb2 + face_emb3 + face_emb4)/4, ...)

johndpope commented 8 months ago

Downstream has some nice illustrations- https://github.com/ZHO-ZHO-ZHO/ComfyUI-InstantID - I played around with thie workflows - but couldn’t get results that I was after. It seems all the examples are based off text concatenation. - somehow interrogating image for words to add to prompt Said another way - what I want is - given an image - can instantid make new images that look like it? Or is this not a suitable use case? I realise t2i adapter colour will best fit here - but then how does instantid shine? Can you create something that looks like this?

(this is the example in ComfyUI-InstantID) Screenshot from 2024-01-25 06-58-10

bent1e commented 8 months ago

Adapt with IP Adapter to control the style of images by inputting ref instead of text prompt.

johndpope commented 8 months ago

Thanks @bent1e - looks like the ref image has been updated. testing now Screenshot from 2024-01-25 13-46-02

marsplant commented 8 months ago

Face swapper is wanted!

Kuvshin8 commented 8 months ago

Samplers? Sde/2M Karras or Euler A? I don't find any mentions about it

P.S. Thanks for your models

brausesa commented 8 months ago

Please make it possible for the local version to specify long tasks for generating images.... For example, so that I could specify one reference image of a face, plus specify several images of poses, plus, for example, select twenty different styles (prompts) and the number of variants (generations) for each style... and after pressing the "submit" button, I could go to sleep... And in the morning look ouput-folder with a hundred results :)

Thanks! You have made an incredible product!

min-star commented 8 months ago

Face swapper is wanted!

InstantID is not only swapper face,did you use it with other ways

ameen-roayan commented 8 months ago

a simple comfy ui implementation would be great.

johndpope commented 8 months ago

Faceswapping is here - https://github.com/InstantID/InstantID/pull/89

sbyebss commented 8 months ago

Although it's already mentioned here, I hope Multi-ID can be added

cqflgl commented 8 months ago

项目很惊艳，非常感谢开源！ SD1.5版本的生态是最好的，如果能支持sd1.5，相信用户数和github星数都会上一个大台阶。

cubiq commented 8 months ago

Multi-ID is kinda possible already (in comfyui) even though a bit complicated at the moment, I'll see if I can find an easier way to do it https://github.com/InstantID/InstantID/issues/81#issuecomment-1937460812

johndpope commented 8 months ago

Im interested to use this technology with music. there was some prior work done with riffusion - but lack luster model support - this could with the help of a new model - reopen box to create styles of music from spectrograms + prompts.