NooblyAI
Ignore v1, v2 is better. NoobAI EPS v1.0 base.
A fine tune of NoobAI-XL 0.75 EPS using a dataset of 390k images.
Produces fun and vibrant images even at low step values and medium resolutions with Euler a sampler.
The idea behind this merge was to make a more coherent version of the awesome Noob 0.75 release, along with a touch of Niji aesthetic.
Recommended negative prompt:
lowres, worst quality, low quality, bad anatomy, bad hands, multiple views, abstract, signature, furry, anthro, bkub, 2koma, 4koma, comic, mangaDescription
FAQ
Comments (6)
The most underestimated model.
I tested about 5 other fine-tuned models based on NoobAI v1 so far. All of them have some kind of overtraining problems. They forget uncommon prompt words and tend to generate the same color and pose. (Edited: Mainly "2girls"/multi-person prompts. I guess maybe because the fine-tuning data is focused on "1girl"?)
This model has a beautiful default stable style, but can still generate flexible colors/poses/uncommon concepts. Lora compatibility and stability are also very good.
Thanks for the awesome checkpoint.
results have been fantastic! this and obsession are my current favs of illustrious/noob so far!
Hi mate! This model is really something. The prompt following and overall image quality is straight phenomenal. I appreciate your input in whole thing. May you please share info about how to tune something like that? Especially how to tune-in new things and not damage already working stuff like a bunch of artist styles? Your previous guide on pony fine-tune was incredibely useful.
Did this become NoobEVO? I'm so confused
Great model. Good colors, does backgrounds well without even needing to prompt for them, and is quite creative too
using this model is still a very pleasing experience, hands down. Such nice textures, colors and backgrounds. Ideal for 1990s-2000s anime especially.