>>106296774
Qwen is doing the same thing. It's hiding faces you guys think it's next gen for whatever reason kek. The Qwen image is not sharp. Chroma can do images with sharp faces in the background (just not perfectly yet), in small crowds, but it's something. I have shown multiple examples of this just one or two threads back
>>106276976
>>106276540
>>106276522

You're not showing something uber impressive that Chroma wouldn't get right with just a LoRA trained. In fact, with prompt engineering look what I can do. The blur makes it an easy thing. What would be impressive are coherent sharp faces in a crowd, which I would admit is a limitation of Chroma. And obviously, in the Qwen example you may have more accurate representation of Japanese streets and shops, or things that happen in the background, but that is also a limitation that can be overcome a good LoRA.