Skip to content
Hugging Face Blog·· 2022-11-07AI score38

A Tuning Guide for Training Stable Diffusion with Diffusers and Dreambooth

Training Stable Diffusion with Dreambooth using Diffusers

AI briefing

When fine-tuning Stable Diffusion with the 🧨 Diffusers Dreambooth script, a low learning rate combined with gradually increasing training steps works best: faces need 800-1200 steps, batch size 2, and LR 1e-6, while objects need 2e-6 with 400 steps.

Open original huggingface.co