search

tencent-ailab / IP-Adapter

The image prompt adapter is designed to enable a pretrained text-to-image diffusion model to generate images with image prompt.
Apache License 2.0
4.5k stars 296 forks source link

sdxl_plus_face result is bad on many image #284

Open akk-123 opened 5 months ago

akk-123 commented 5 months ago

I test the sdxl_plus_face on some img, all param is same, only input image is different. I found that on some img(beautiful woman), it can get good result; while on other img, the result is bad

for example prompt: photo of a beautiful girl wearing casual shirt in a garden

image image

image image

xiaohu2015 commented 5 months ago

hi, you can use good text prompts or good base models

akk-123 commented 5 months ago

I try different models and prompt, but can't get good result

xiaohu2015 commented 5 months ago

I try different models and prompt, but can't get good result

lower the scale of IP-Adapter? you can also try to combine other face adapter models

akk-123 commented 5 months ago

@xiaohu2015 will release sdxl-faceid-portait?

xiaohu2015 commented 5 months ago

@xiaohu2015 will release sdxl-faceid-portait?

I will open if I get good results.