Retrain of Not The True world (SD1.5/SDXL)
You can prompt for specific style in the model, but if you choose to load this Lora with any prompt in txt2img or img2img, you will benefit for enhanced structural details & alignment.
V1-lite = 2-16, Pruned dataset, trained with full repeats (20) for 8 Epochs, cosine, Prodigy
Captions for pruned dataset -> in the Training data download ->
V2 = 2-16, full dataset, trained with half repeats (10) for 11 Epochs, cosine, Prodigy
Captions for full dataset here
V3 = V2 retrain in 4-16, full dataset, 10 repeats for 9 Epochs, cosine, Prodigy
V3-lite = V1 retrained in 4-16
V4-lite = V1 retrained in 8-16
more information be added...
V5-lite = OG data with all new Captions (fast flux train)
V55-lite = same dataset as V5-lite, with Full Training Config.