2209.12330 Personalizing Text-to-Image Generation via Aesthetic Gradients
Daily Image Generation AI (October 22, 2022)｜Yamakaz｜note
-
f627e40af195e35a860d28fe754d5d2eee4b2f|twcon^s1_c10&ref_url=https%3A%2F%2Fnote.com%2Fyamkaz%2Fn%2Fn55c3ac365f34 @MutedGrass:. To learn aesthetic images embedding, just dump the trend-aligned images into a folder, go to Train/Create aesthetic images embedding in AUTOMATIC1111, specify the name and folder, and press the button. No cropping or text is required. It takes no time at all.
@MutedGrass: Aesthetic Gradients effect comparison. The first one is “hmmm”, but the second one is significantly more effective when the aesthetics learning rate is increased. However, it significantly weakens the other elements of the original prompt, so it is troublesome to adjust the balance.
- (Images are NSFW)
- Hmmm, I haven’t seen much paint in the generation so far, but the control at the prompt doesn’t work very well.
- I think they’re still fixing bugs and stuff, so I’ll hold off on that for now.
This page is auto-translated from [/nishio/Aesthetic gradients](https://scrapbox.io/nishio/Aesthetic gradients) 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.