Some algorithms really mess up faces on otherwise great results. I've found this tool to be pretty snappy when it comes to post-generation touch-up: https://arc.tencent.com/en/ai-demos/faceRestoration
Example, generated with DALL·E 3:
Prompt: "A photo of a couple at a movie theater, one of the women is pregnant and very far along. She does not touch her belly with her hands. Her belly is so large and spherical that it almost touches the screen. She also looks bored."
It spits out an image that's twice the size of the original, but it also smooths out some of the finer details.
Since all I care about is fixing the face, I paste the new face on top of the old one using a photo editor, and I end up with something like this:
If the face correction comes out a little too yassified, you can blend it with the original until it looks right. HTH!
Example, generated with DALL·E 3:
Prompt: "A photo of a couple at a movie theater, one of the women is pregnant and very far along. She does not touch her belly with her hands. Her belly is so large and spherical that it almost touches the screen. She also looks bored."
It spits out an image that's twice the size of the original, but it also smooths out some of the finer details.
Since all I care about is fixing the face, I paste the new face on top of the old one using a photo editor, and I end up with something like this:
If the face correction comes out a little too yassified, you can blend it with the original until it looks right. HTH!
