Try [Imagic

  • image

Colab: https://colab.research.google.com/github/ShivamShrirao/diffusers/blob/main/examples/imagic/Imagic_Stable_Diffusion.ipynb#scrollTo=K6xoHWSsbcS3

Input Image(src, CC-BY) image

Prompt: “A girl with flower.” imageimageimageimage Oh, if you only specify GIRL, then cat ears and hair color features that are not specified will be ignored.

Original Prompt + “with flower” imageimageimageimage This time, only the background color has changed.

Is Gradio going to bring up the GUI, that’s great.

  • Prompt strength was set to 0.9 by default.
  • I’ll chop it up because it didn’t change much when I set it to 1, and when I set it to 2, it broke down considerably.

1.5 imageimageimageimage 1.75 imageimageimageimage 2.0 imageimageimageimage

imageimage I’d say this or something like this retains the characteristics of the original image, but with flowers as indicated.

  • I was expecting a single flower decoration.
  • I never thought I’d interpret ear hairs as flowers.
  • Well, but I might decorate it that way, I guess.

Interesting points

  • The original image was created by NovelAI
  • Imagic itself is based on Stable Diffusion 1.4
  • As a result, it is possible to express the animated picture in a way that is not possible on NovelAI’s service, which should be quite difficult with Stable Diffusion by itself.

PS

  • NovelAI img2img “flower” strength 0.7 noise 0.2
    • image
  • noise 0
    • image
  • stength 0.6
    • image
    • This is close to what I was hoping for.
    • I noticed I left “{{3d}}” in the negative prompt, so I removed it, but it doesn’t make much difference.
      • imageimage
      • The second is the SEED difference

This page is auto-translated from [/nishio/Imagic 2022-10-31](https://scrapbox.io/nishio/Imagic 2022-10-31) using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.