I hate it when this happens...
For the other side of the wall... https://civitai.com/models/8817/trapped-in-wall-behind
Weights of 0.7 and 0.8 works pretty well for me. I wasn't able to combine it with character embeddings very well. I didn't try all that hard though.
On realistic models, you will notice how the hands have the wrong perspective. I was looking at the dataset and noticed that the hands are wrong there too! So, it trained correctly and is recreating things as they model was taught to. I'll think about how to correct this in the future.
Small improvements. Works better on realistic models. Hands are a tiny bit more consistent.
Is a little over trained. A weight of 0.7 or 0.8 works well.
Examples are the same seeds and nearly the same prompts, but using the new version!
Some of the negative prompts are different. I got half way through before realizing that. I don't want to do it again...
Files
Trapped_In_WallV2.safetensors