search

rinongal / textual_inversion

MIT License
2.87k stars 278 forks source link

A question about picture generation #159

Open 2629630207 opened 10 months ago

2629630207 commented 10 months ago

I'm very satisfied with the images generated using my face as the training dataset. However, is it possible to make the generated images perform additional actions, such as "* play football in the playground"? I used similar sentences as prompts, but in the end, only faces were generated.

rinongal commented 10 months ago

Textual inversion has a noticeable tradeoff between identity and editability. You can tune parameters to get a better point on this tradeoff, but honestly at this point my suggestion is to look at newer methods like LoRA or Neural Textual Inversion.