<pre><code>After going through twists and turns, I started doing it from noon to evening, and continued after meals. Until now, I have tested it repeatedly.

The background color of the training set is green, which can also be said to be the color of the wall. I understood it as the background color at the time. However, I did it several times and thought it was fitting, so I deleted the archive and changed the green color in the training set. The wall was buckled and turned into a white wall, and then the magic happened. I found that the color of the wall could not be changed after I took out the picture. I also kept prompting words for the red background and various background colors, but I could not change the color of the white background. , I thought it was a good fit, but with the weight of 1, the naked girl can assume various poses, and all hair colors and hairstyles can be transformed. Then I applied loras in other costumes, and found that it was a bit over-simulated. If two loras are used at the same time, they need to be transformed. The weight of this lora was reduced to 0.7, and the clothing was reduced to 0.8, but the background was still white. I went out for a walk, came back and thought about it, and tried writing a red wall. As a result, the AI ​​​​interpreted it as a wall, not a background. Now it's ok. For testing, you can add "windows, doors, open doors, some scenery outside the window," and the main subject is still hanging roses or roses in the background.
Recommended weight is 0.9</code></pre>

naked leaves

<pre><code>You can add more words, windows, blinds, doors, the scenery outside the window, indoors, outdoors, in short, you can think of it. I want to see the return picture, where is the limit of this lora. Thank you.

Recommended weight is 0.7-0.9</code></pre>

Files with SHA256: 13c61e692fb98a8a131b5e160a536052efeac3c6c4c46a660b8cd3dc3c16e77e