Hi, This model tuned on IllumiYume XL V2. Which Trained on Danbroou 2024 dataset, it can make most of new character too. I just improved
Shadow and Lighting
Prompt Accuracy (Highly improved)
Effect and Style stability
Super Adaptive with any lora
Feel free to try my models on Tensor.Art
V2 :- Improved prompt understanding, Hands and Artist style much accurate, skin quality improved.
Use Adetailer For Max Details
Description
V2:- Hands and artist Style much accurate
FAQ
Comments (15)
version 2.0 is more realistic compared to version 1.0
yes... i enhance skin and details for better look better
I am really impressed, this checkpoint is incredible. Thank you very much for sharing, I am really having a lot of fun because of its stability!
I am glad you like this... I am working on more models mostly gonna be TensorArt exclusive. You check my models there too. 🙏
@nukeai1106 I am still using this checkpoint and I absolutely love it. Do you think you'll ever upload here your anime screencap models I saw on TensorArt? They are so gorgeous!
@dunkelheitdx770 glad to hear that... And sorry that style is TA exclusive. They paid for Full frenchise 🙏
Very good checkpoint, it has become my favorite. Thank you.
Thank you so much... I will gonna make it more better in next version. 👍
Perfect, Thank you!
Excellent, you will upload more models??
well i am working on new base like Anima for Yume :)
this is probably one of the most underrated model you had. its better than tensor exclusive model ver that you did, and almost equal to the other exclusive one you made on seaart after thousands of testing generation, in term of consistency and stability.
the only cons is just it requires Adetailer as prerequisite. but other than that, it beats the other models hand down. (some other model doesnt require this), good tradeoff for amazing anatomy quality though. I assume thats why some people says it lacks facial quality, as it relies completely on adetailer to do its job. So if they generated without adetailer it will feel bad quality.
one of the pro of this model is that because its actually originated from vpred+illustrious, so the loras are interchangeable. I was surprised to see my vpred lora works, and il lora works, i can even combine both. There is a caveat though, only Loras trained on ILXL 1.1 above 9000 steps to be exact.
But since training over 10000 is normal for my side, it actually feels great using this model. it reminds me a lot of AnimagineXL 3.1 quality, with level of prompt thats far superior.
well yes thats also on V-pred which give it extra prompt and character understanding boost... i was planning to make more style and version on Yume tune but user feedback quite low on civitai as compare to TA user keep asking and demand what they want. but in civitai its quite feel off.