search

rinongal / textual_inversion

MIT License
2.9k stars 279 forks source link

Text guidance #123

Open Anastasia0411 opened 1 year ago

Anastasia0411 commented 1 year ago

Hello. I've faced the following issue: I use stable diffusion model (checkpoint sd-v1-4 from https://github.com/CompVis/stable-diffusion) to introduce this concept image Unfortunately, during inference time the model ignores all additional words in my prompt. It generates just this teapot and nothing else. For example for the prompt "a * is floating in the sea", I got image This teapot looks good, but there is no information about sea on this image. I tried different guidance scales with no positive effect. Has someone else faced this problem as well? Thank you

rinongal commented 1 year ago

Sorry for the late response.

How many vectors are you using to represent your concept? And what learning rate? You might be overfitting the concept.