Is this model good? Definitely in top 5% among Illustrious models here. It don't follow prompts strictly, but overall results look diverse and unique in a specific way. Still, I would try to improve it in future.
I don't have intention to cherry pick/upscale/animate example images - that's how it is in first try.
recommended beginning of positive prompt (it just works):
masterpiece, masterful,recommended negative prompt:
(ai generated:0.7), ai fingers, grayscale, sketch, very bad quality, simple background,
If you don't care about speed:
dpm++ 2s ancestral / beta / 40 steps / CFG 4.0-5.0 - the best quality in average (examples)
or
dpm++ 2m sde/ sgm uniform / 80 steps / CFG 6.5 (examples)
For 25 steps:
Euler + simple = more diversity
Euler Ancestral + simple = more stability (good for upscale)
CFG: 5-10 (depends on complexity of prompt - use more for short prompts)
Sometimes this works very well for complicated prompts ("burned" images for simple prompts):
Euler_cfg++ / beta / CFG 2.5 / 25 steps - For more steps you need lower CFG - CFG 2.0, 35 steps is another alternative, for example.
The model don't have a specific basic style, so use tags/descriptions as much as you can to improve quality.
The main biggest component in merge is basic Illustrious 2 model. It's not a simple merge - it was merge by blocks, so it's a lot of work (hundreds of hours of manual testing and merging when results were improved in average). I have generated almost 20 000 images only to test and compare different combinations of weights. Current v1 is 7th version I have. A lot of different models were used in process (mostly close to basic illustrious versions or good universal finetunes).