Trained on a normal and natural dataset. For natural lighting, textures, details...
The dataset only has high resolution images. Zero AI image. So you can get texture and details beyond pixel level. Instead of a weird smooth feeling. FYI, all cover images directly come from a1111, no upscale, no inpainting. If you think they are, then this LoRA did its job well.
Recent updates:
(5/19/2025): Recommend to use illus version of this LoRA even if your base model is NoobAI.
Ironically, last time I update. I recommended to use NoobAI version, because I found 95% base models are NoobAI.
Well, "you should use LoRA on the base model that it was trained on", that is still true... But may not ideal for natural detail models such as this LoRA.
Long story short, I grabbed a 5K dataset and did some big trainings last week, on illus and nbep. I noticed that nbep "refuses" to learn details, even if the training settings are identical and nbep was trained with more steps. So illus versions LoRA always had better effects on both illus and nbep model.
I also tried to train on the new Illustrious 2.0, the result is: It is nearly un-trainable, much worse than NoobAi.
So my assumption is that it is still the "overfitting" problem, the more a model trained from original SDXL, the more it forgets. And the hard for it to learn things. (?)
Anyway, remember to use illus version. For demonstration, all illus01 v1.152 cover images are from NoobAI e-pred v1.1.