Dataset cutoff: 2025/07/15

Continued training from v2.0

This version now includes training on an updated personal aesthetic finetuning dataset (masterpieces), and some recent characters, outfits, and styles:

• Training images: 1,004
• Regularization images: 314 (Generated from v2.0)
• Optimizer: Adafactor
• Training precision: Full-fat fp32
• Batch size: 4
• U-Net LR: 6e-6
• TE LR: 2e-6
• Epochs: 50
• Steps: 25,950Now totaling of 378,550 steps trained from base (~370 GPU hours at 3.52s/it)

List of new series/characters trained:

anime:

• dr. stone
• gachiakuta
• kaijuu no.8
• kaoru hana wa rin to saku
• sono bisque doll wa koi wo suru
• yofukashi no uta

virtual youtubers:

• sameko saba
• hololive holonatsu paradise (outfits)