My first mix in illustrious category aiming to bring more details and flexibility to animated style of this type of models. Various results can be achieved - from animated to semi-realistic style depending on the prompt.
As always my gratitude and due respect to the creators of original models that are used in the mix:
https://civarchive.com/models/827184/wai-nsfw-illustrious-sdxl?modelVersionId=1490781
https://civarchive.com/models/1231009/be-mix-illustrious?modelVersionId=1387129
https://civarchive.com/models/959736/rillusm-realistic-il?modelVersionId=1448475
https://civarchive.com/models/1412760/lunarcherrymix-illustrious?modelVersionId=1637795
https://civarchive.com/models/1695253/smooth-mix-illustrious2-noobai
https://civarchive.com/models/1318945?modelVersionId=2319122
Version 6.0: Experimental formula with addition of new elements to the mix for future development.
Version 5.0: Update with newer versions of the components.
Version 4.0: Return to older (v2.1) formula with improvements in composition and prompt adherence.
Version 3.0: Experimental mix with a little shift to more anime style and more vivid palette.
Version 2.1: Improved stability, composition and faces.
Version 2: Cranking up overall details and contrast. Colors look more vivid too. Needs further testing. Has more of its "own identity" compared to version 1.0.
Version 1: Overall very decent prompt adherence and anatomy. Stylisticaly close to anime-semirealism. High flexibility: some combinations of LORAs give very interesting results (you can find examples in the images).
Recommended setitings:
Sampling method: Euler a, DPM++ 2M, DPM++ 2M SDE, DPM++ 2S a, DPM2 a, DEIS
Schedule type: Karras, SGM Uniform
Sampling steps: 20-50
CFG Scale: 3-7
Highres.fix + Adetailer for face are recommended
Description
New base for future development
FAQ
Comments (14)
Will v.8 be available for onsite gen鈥檌ng soon? And what are the updated features? Thank you! 馃檹
Yes, next week both v7 and v8 will be available for onsite generation for comparison. After that I will bid only for v8 (depending on a demand). v8 has a significantly different models at its core (because some models that were used for it before are no longer updated by their creators) and it is toned down in detalisation alittle. You can expect better LORAs compatibility because of that.
@Axelros聽Will I have the ability to bid for v.7 going forward?
@Aniartcanvas聽Yes, you can bid for any checkpoint in auction
What changed in v8?
v8 has a significantly different models at its core (because some models that were used for it before are no longer updated by their creators) and it is toned down in detalisation alittle. You can expect better LORAs compatibility because of that.
@Axelros聽Thanks! If I don't use LORA will I notice an improvement, same, worse?
One thing I notice from v7 is that scenes/skin tones appear brighter in contrast. Not washed out, but closer to that. v8 at same CFG 7 looks a little darker with better shading detail that I thought looked better. Not sure if there's a lever to control this.
@voidpointer1It is not as detailed and bright as v7, but has a greater flexibility in styles and prompt adherence, plus a more up to date character understanding聽
@Axelros聽sorry I had my comment backwards: v7 looked slightly darker with better shading, v8 looks brighter, less overall detail/contrast. Do you know of any settings or second pass workflows that might go back to that darker contrast?
@voidpointer1聽I'm afraid you can increase contrast only with LORAs
@Axelros聽Do you recommend the lora approach + stick with v8? Or go back to v7?
@voidpointer1聽It is your decision to make. Experiment and find out what you like more...
THE BEST MODEL IVE USED FOR IMG TO IMG. VERY CLEAN AND DETAILED.. AMAZING WORK!!