/vt/ - Virtual Youtubers

>>72636190
The cosplay LoRA was not the solution, but block merging was. I was able to identify the block that resulted in a mess when combining pony lyco with animagine (block 6) and I am now able to block-merge up to 60% pony unet without the model breaking (75% works as well but worse/requires more fiddling).
It's still not too great at hands if they're not in-focus and I didn't get to play with it too much outside of testing. Maybe ninomae ver. tomorrow.
https://files.catbox.moe/9fo5m8.jpg

Also, if anyone is as lazy as I am and you use Grabber to right click an image from danbooru then 'copy tags' and get annoyed at all the 'blue_dress green_shoes' tagging not being parsed for SD, I made (copy-pasted from ChatGPT &debugged for myself) a python GUI parser with clipboard function. You can use it to create a prompt with one hand if you're having lunch or vaping with the other. I'm not implying any other use case.
https://github.com/CCC-anon/CCC-Scripts/tree/main?tab=readme-ov-file#tagsuipy