/vt/ - Virtual Youtubers

>>46993295
>without them clobbering their tags
I wouldn't say that you are merging tags per say, since that would imply that you are merging the clip encoder rather than the model itself. You are merging the UNET Blocks, which is incredibly abstract. (Although if you also want to fuck around with merging the clip encoder, then mucking around with base_alpha will control how much of model_Bs Clip encoder is merged with Model A's.)
>Let's say I want my holo model to take some cyber robot concepts from some model off Civitai without my chuubas tags getting wiped, which preset should I pick?
There isn't one preset to rule them all. You can use a preset as a starting point and go from there.
>What are the IN and OUT sliders mean? I tried reading the manual but shit is like a textbook, it says some of them control art concepts, lighting, shapes etc..
Okay, so the IN sliders deal with the IN blocks in your UNET, these downsample the image while adding concepts until it reaches MID, at which point the UNET starts upscaling the image and adding concepts again. The IN and OUT sliders do the same thing, they just refer to different locations in the UNET. A good rule of thumb is that concepts, and compositions are going to be primarily in the IN Blocks, while they control texture and colors secondarily, the OUT Blocks are primarily responsible for Colors and textures, with Concepts and composition coming secondarily. Pic Related gives a ~~~~VERY~~~~ approximate description of what could be in which block, however every model is different and concepts may be else where due to different weights of the prompts and tensor deviations.
>What preset did you use? I don't mind doing some brute force but I'd like a good starting point.
I didn't use any presets, I used a black box optimizer in order to find good merge values. However, if you look here: >>46992178 , you should be able to gleam some critical information about what might be where for the HLL model at the very least. The darkest red areas are areas where the HLL blocks completely replace the BBW blocks. If you are merging an HLL model into another model keep a close eye (which is to say, make these blocks closer to 1) on IN00, IN01, IN04,IN05,IN06,IN08, and OUT02, OUT07,OUT09, and OUT10. As the anatomy concept you are talking about falls in the areas that are more blue in the post I mentioned above. I would personally start with the flat 25, and just screw around from there if you are going to do it manually.