This is a Bra V6 Fixer.

The Bra V6 is a fantastic model, but getting a good handle on the Brav6 prompts could be a bit of a challenge.

The Bra V6 fixer helps the prompts work a little more as intended, and the LoRAs come out a bit nicer.

Recipe

procedure: almost the same method applied for "chillout NFSW fixer"

1. try to interpolate input_blocks of Bra V6 to chilloutmix to make LoRA friendly

2. try to interpolate output_blocks of Bra V6 to dreamshape v8

3. make a checkpoint

4. make LoRA using Supermerger (by https://github.com/hako-mikan)/ make Lycoris/Locon version using LyCORIS extract command line tool foundat LyCORIS github project (by https://twitter.com/KBlueleaf)

Please see the following article.

How to make your checkpoint LoRA friendly?

Usage

• Lora version v1:

  • "<lora:Brav6fix_v1:1>" : LoRA weight ratio could be 0.5~1.5

• Lycoris/Locon version v1:

  • "<lyco:Brav6fix_locon_v1:0.9>" : Lyco weight ratio could be 0.5~1.5

  • for Lycoris version : download "locon v1" version and copy into stable-diffusion-webui\models\LyCORISsubdir. (it is better to rename downloaded file correctly if the filename is not "Brav6fix_locon_v1.safetensors")

  • from webui 1.5.x(?) you can use <lora:Brav6fix_locon_v1:0.9> without extension. (in this case, you have to copy lyco/locon file into models/LoRA

    • but it seems a bug exists that makes Lycoris not work properly.

License

The same License of Bra V6