Z-Image Turbo Furry Edition

Trained on my whole entire collection of 28451 images. I have no clue how TF this worked, Qwen complained at 700 images while this turbo one just trucked through it. Though I did use a lora trainer called AI Toolkit which IMO feels better than OneTrainer but it's missing a lot of stuff. BTW this starts to show furry stuff at like 1200 steps.

No promises but I'm doing a custom one which less images.

This was trained using the adapter.

Who needs to make a model when Lora do trick.

Trained Resolution (512x512, 768x768, 1024x1024)

Generation time 15 hours 54 minutes 1 second (RTX 5090 limited to 400w or 67%)

Images 28451

No quantization on AI Toolkit

Rank 128

64GB of RAM

Captioning (Taggui with wd 3 large) did I check every caption, no, you crazy

Pre-Test

this is just the end of training, I will try my best to make it better

TRIGGER WORD

furry