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
4.46k stars 289 forks source link

Is training on small datasets feasible #383

Open Qianjie007 opened 2 weeks ago

Qianjie007 commented 2 weeks ago

Hello dear author. I used a small dataset of 200 (clothing texture) image text pairs for training, and wanted to generate textured images by inputting regular images, but the effect was very poor. The generated image is not only unrelated to the original image, but also unrelated to the training set. May I ask, what efforts should I make next? 图片1

xiaohu2015 commented 1 week ago

maybe test with controlnet?