Description
- Currently, the model still has many shortcomings, but it can still produce good images of basic portraits and slightly more distant half-length portraits. Exploring the work of training the SDXL Asian female model to output high-quality images is not easy, everyone can join in. I hope to work together with all the experts to improve or surpass the representation of SDXL characters and achieve effects beyond SD1.5.
- 1. Please describe in natural language. Later, I tested it, and it can also output good results with simple tags.
- For example: xxmix_girl, a woman with long hair wearing a white shirt and a pair of earrings on her neck and shoulder, with a white background,
- Opening line: xxmix_girl, a woman ____________________________ or xxmix_girl, 1girl____________________________ (1girl feels younger, a woman appears older)
- Currently usable prompt: xxmix_girl, 1girl, fluffy short hair, dark theme, blonde hair, messy hair, film grain, fog, black theme, Tyndall.
- Other prompts can refer to image information. I don't have too many styles, but it is recommended that prompts should not be too long, as any word might affect the image output.
- Reverse prompting words: worst quality, low quality, illustration, 3d, 2d, painting, cartoons, sketch, tooth, open mouth, (It is recommended to follow mine).
- 2. For now, the model still recommends portraits or half-length portraits. Of course, with the high-resolution restoration and afterdetailer extension plugins, full-body is also possible, but prompt words are very important, you never know which prompt word might hit the mark on the Western faces 0.0.