Talk:AI Art Generation Handbook/Training

Possible Dreambooth Training Parameter - In Progress
d8ahazard / sd_dreambooth_extension: 926ae204

Vladmandic Auto Fork from Auto1111 Version:  22090d7b

--Hardware-- 16GB VRAM NVidia RTX3080 Usage: 14.5GB VR ---

Dataset:

Images : 44 different pictures (512X512px)

---

If not mentioned, it is default values.

Training Steps Per Image : 175

Batch Size : 4

Learning rate : 0.00000125

Use EMA : True

Optimizer: 8bit AdamW

Mixed Precision: bf16

Memory Attention: flash_attention

Weight Decay: 0.01

Time taken: ~ 1hour

MrAlanKoh (discuss • contribs) 04:47, 5 May 2023 (UTC)

Captioning Link
https://github.com/baaivision/CapsFusion MrAlanKoh (discuss • contribs) 10:49, 15 June 2024 (UTC)