译文
/ Text
Hong Kong style red lip makeup & plain water makeup
After continuous attempts, I actually found a point
GPT's control over makeup is still too big a gap compared to bananas. For GPT, relying solely on text prompts to generate images, even if all the anchor points are the same, the resulting images will be different. If you really want a makeup look that satisfies you, here are my personal recommendations:
1. First, use a draw card method to select a satisfactory makeup look, and then use the reference method for anchoring. This effect will be better and more stable than that of pure text.
2. When you have a desired makeup look, it is recommended to first reverse decompose the makeup, and then apply it using a combination of reference images and text locking. The same makeup look can be completely different in terms of facial structure and lighting effects.