The effect is as shown in the picture, much better than direct color replacement. The downside is that it requires some luck or PS.
If there is no response when generating the image, it is recommended to re-enter the color's word.
For color replacement, you need to input the corresponding color's English name, such as yellow, red, green, etc. Inputting a color code has little effect or no response.
For the target area, you need to input prompt words indicating the clothing parts you want to change, such as clothes, trousers, dresses, hair, etc., in English (multiple inputs can be separated by commas, e.g., clothes, trousers. The software's cutout may sometimes include or exclude too much. In such cases, either adjust the prompt words to be more specific or use a different image). The default in the workflow is clothes. Common prompt words are as follows:
衣服 clothes
裤子 Trousers
鞋子 Shoes
连衣裙 Dresses
眼镜 Glasses
The first run takes about 1 minute and 40 seconds, and subsequent runs take about 1 minute. The output is two combined images; the image on the right has more details added. Generally speaking, the right-side image is better, but for certain colors, the left-side effect may be better.
It responds well to simple patterns on clothes and trousers but is less responsive to complex patterns on lingerie.