Lativi art style model, recommend to use NoobAI-XL generate.
Dataset sourced from the author's Pixiv and X (Twitter).
If this model causes any inconvenience to the original author, it will be removed upon request.
Some recommend:
Lora weight : 0.7 ~ 1
Steps : 20 ~ 28
CFG : 3.0 ~ 5.5
Sampler : Euler
Scheduler : simple / sgm_uniform / karras
Description
FAQ
Comments (3)
Hi, I wanted to say that LORA looks amazing, but I confess I'm not familiar with the NoobAI-XL; I think the WAI-illustrious-SDXL is almost universally accepted. So, could you specify which specific NoobAI-XL model you used for each example image, please? I'd like to test that model; it looks good.
Edit: Oh, and if you can and still have it saved, could you post the positive and negative prompts you used to create each of the 4 example images? The poster can edit those details for each image. Please, this will definitely help improve your LORA, which is already amazing!
Every example image was created with the epsilon-pred v1.1 model. They retain full metadata, so you can drag the downloaded images directly into ComfyUI to load the workflow and prompts.
To be fair, both noobai and illustrous are quite outdated as sdxl models. But, sure, there are almost no decent alternative to it that can run on consumer gpus, so we are stuck with illustrous and noobai which are both widespread. I have an eye for lumina models. they are incredibly good at perceiving prompts since it uses llm as clip. but it lacks community support. hope it will get more recognition as next gen model.