Comment by GaggiX

>an open-source Stable Diffusion, which is a basis for multiple projects, including commercial Midjourney.

source? The model behind Midjourney looks quite different from Stable Diffusion, to me it looks like a model conditioned to image embeddings (like Karlo or Dalle 2) instead of text embeddings, also I don't see any mention of the Creative ML OpenRAIL-M license on their website, when they experimented with SD in August they added it but it has not been there for months, To me it is clear that MJ has trained a new diffusion model from scratch, it is also noticeable by the difference in performance, even for a model trained in AI image recognition it struggles with images generated by Nijijourney (MJ but finetuned for anime) while it has an easy time with anime images generated by SD and its finetuned models [1], which means that the distribution of images is very different.

[1]: https://huggingface.co/spaces/saltacc/anime-ai-detect