The Qwen 2512 took about a week. because it kept producing images that looked like they were printed on embossed paper towels. In the end I used Musbi-tuner and it worked a treat. I forgot the trigger word though, but it doesn't seem to matter. Just use something like 'A photograph of a young East Asian Woman in her early 20s." The Moss and Qwen V 2.1 versions were done on AI-Toolkit, and created a reasonable LoRA in one hit.




