Upscale your character's unique face! This workflow is designed as the second part to generating diverse, random faces, based on randomly generated names. By doing this, we sidestep the use of celebrity images and names, ensuring a pool of original names and faces for your characters, free from ethical dilemmas.

How to use it:

0. Works best if you used my Face Find SDXL workflow and have an image associated with a name  https://openart.ai/workflows/-/-/OsArpYrUDTAuqYsR0nKc#comment-2RTMLlUO4lYhQ8AbifqA

1. Load a face image

2. Enter your character's name (works best if it was the name of the character used from the random generation workflow)

3. Choose your model and upscale multiplier (I recommend JuggernautXL and upscale 1.5x)

4. Run it!

Tips:

1. Keep the checkpoint and character name as the ones used to generate the face in the FaceFindXL workflow, this helps stabilize the upscale.

2. You can experiment, but I generally don't exceed 1.5x the upscale size. For the model upscale, the eyes get weird, and for latent upscale I usually end up with weird face lumps. If you want to go larger, I recommend passing the result back in as the input image and upscaling by 1.5 again. Also, this step is just to add a little detail to the face. But it will get downscaled for use in the next phase of my workflow steps, so it doesn't have to be huge.

3. Try out the different upscale options and see which you like best. My favorite is the Neural NNLatentUpscale but note that you need to download the custom node manually because it isn't in the Manager database.  https://github.com/Ttl/ComfyUi_NNLatentUpscale

Next Steps (How I use this in my overall workflow):

1. Use the face image in a face replace workflow, my preference is IPAdapter, but FaceDetailer and Reactor work well too

Notes:

-- Sometimes I end up liking the original image better than the upscale. So your mileage may vary and you can skip this step entirely sometimes =)

v3 Update 3-2-24 - Added self-attention guidance and removed ComfyRoll Latent selector which was causing errors.