V2 - charges in pushing more prompt adhearance and blending in another custom finetune
Experimental Release
-- Explanation - Juggernaut-Z-Image came out and I wasn't happy with the results, but the dataset seemed solid, so I went to work in adjusting the block layers. I also wanted a fp32 so I can convert it to nvfp4 for a high quality nvfp4 file. So I took the original Z-image-base fp32 and blocked adjusted it to match the new model. I took the two newly created block adjusted files and smooshed em together at a 70/30 with Z-image-base at 70 and then did the converts. I would of have done ggufs, but I do not know of that method for these types of models. The datasets stay the same for both models, but the model block layers are adjusted. The goal was to push details and prompt adhearance to push ZiB to do better. Success? I'm happy with it, ymmv. Thanks to the great work of everyone in producing these great models. Thanks to those that produce the great tools to make things happen. I don't want to take credit away from these great people.
-- Examples - are provided and each one should have it's workflow, which are all the same and I uploaded a .json file of the workflow.
