Bismuth (Experimental)版本v0.0000001 (ID: 144500)

!!! This is a mistake /failed experiment! It will not produce sensible images on its own. !!!

This is more like a toy.

What you can do with it is use it as base model and then use a real model as refiner setting the "Switch at" to something around 0.2-0.4 (ancestral samplers, like "Euler a", can deal with higher values, deterministic ones, like "Euler", need lower values) to get really random output from exceptionally short prompts like "girl, running". Or you can merge the model with some other model to get results that should look like a mix between your_model and toonify_v10 and darksun_v40. Best check out the showcase images attached.

This an experiment of myself to use the inpainting merge (sd-v1.5-inpainting + (custom_model- sd-v1.5)), described here, with other models.

This model basically is this: toonify_v10 + (darksun_v40 - v1-5-pruned-emaonly) * 1

I thought this might add whatever style differences exist between toonify_v10 and the default sd-v1.5 to darksun_v40. But what I got is something that holds the "essence" of toonify_v10 + darksun_v40 without the capability of the base stable diffusion model to produce sensible images.

It won't generate anything you tell it to, but rather some weird abstract but still "complete" mess. So no random noise but rather something else. On it's own it can generate pretty colors, for some unknown reason small mushrooms and every now and then a sad or afraid chibi face.

描述:

训练词语: