This works similar to the Dalle 2 image variations, the model takes the image embeddings to is conditioned the generation. The normal Stable Diffusion is condi

This works similar to the Dalle 2 image variations, the model takes the image embeddings to is conditioned the generation. The normal Stable Diffusion is conditioned on text embeddings, so to make image variations it was necessary to finetuned the model to accept image embeddings instead. There is also a demo for a limited time https://47725.gradio.app/ https://47725.gradio.app/