Snwy fine-tunes Stable Diffusion on 50,000 DALL-E 2 images
Engineer snwy generated approximately 50,000 images from DALL-E 2 with a diverse prompt corpus before the model shut down this month. The resulting dataset supported a full fine-tune of Stable Diffusion 1.5. Side-by-side comparisons show the fine-tuned model matching DALL-E 2 quality and style, unlike the base model. One example set depicts a chocolate chip cookie rendered with added cosmic elements in the fine-tuned version.
#1612xlr8harder@XLR8HARDERIt's important to do this kind of thing. It will be of interest in the future.
It would be better for labs to release weights of earlier deprecated models, but of course most will never do that.
snwy@snwy_mebefore DALLE-2 was shut down this month, i generated around ~50,000 images from a diverse prompt corpus and did a full fine-tune of SD1.5 on it. the result feels nearly identical to DALLE-2 (left: base SD1.5, right: fine-tuned)