Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Recreating Workflow #42

Closed
skyler14 opened this issue Apr 26, 2024 · 1 comment
Closed

Recreating Workflow #42

skyler14 opened this issue Apr 26, 2024 · 1 comment

Comments

@skyler14
Copy link

Thanks for making this. I really like this project but unfortunately I am doing things with the newer native implementation of InstantID in comfyUI, and cant afford to load two full sized models in my workflows. Can you describe a bit more about how this workflow works and how it can be recreated as nodes.

Is this basically just masking and inpainting, are we doing an innswapper faceswap first, then an img2img instantid-sdxl noising/denoising inference. A mix of both? The fidelity on this swap can be great so it'd be nice to be able to recreate the results.

@nosiu
Copy link
Owner

nosiu commented Apr 27, 2024

This should be a technically still valid but outdated explanation:
instantX-research/InstantID#89

NO innswapper faceswap is used. ONLY instantId

Something like this is happening:

  1. mask -> find the face -> cut out and resize the masked region to make it bigger (we will work on this image)

  2. prepare control_image from step 1.

  3. load face embed

  4. perform inpaint with: your model + instantId + controller_net + control_image from step 2 + resized image from step 1 and face embed from step 3.

  5. resize the resulting image to its original width and height and try to paste it back into the original image by using blur and magic around the edges.

@nosiu nosiu closed this as completed Oct 20, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants