Installation
Set up sd-scripts on Windows or Linux with step-by-step instructions.
Quickstart
Train your first LoRA model in minutes with a minimal working example.
Dataset Preparation
Organize images, write captions, and configure dataset TOML files.
LoRA Training
Learn how to train LoRA adapters across all supported model architectures.
Supported models
sd-scripts supports training on all major open-source image generation models:Stable Diffusion 1.x/2.x
The original SD models with LoRA, fine-tuning, Textual Inversion, and ControlNet support.
SDXL
High-resolution 1024px generation with advanced LoRA and ControlNet training.
FLUX.1
Transformer-based architecture with dual text encoders (CLIP-L + T5-XXL).
SD3 / SD3.5
Multimodal Diffusion Transformer (MMDiT) with three text encoders.
LUMINA
Next-generation DiT-based model with LoRA and fine-tuning support.
HunyuanImage
HunyuanImage 2.1 with LoRA training support.
Anima
MiniTrainDIT architecture with Qwen3 text encoder and LoRA / fine-tuning support.
Training methods
LoRA
Low-rank adaptation — fast, lightweight, and compatible with all supported models.
Fine-tuning
Full model fine-tuning (DreamBooth-style) for deeper style and concept learning.
Textual Inversion
Train new embedding tokens for SD and SDXL models.
ControlNet
Train spatial conditioning networks for precise control over image generation.
LECO
Erase or enhance concepts from a model using only text prompts — no images needed.
LoHa / LoKr
LyCORIS-based alternatives to LoRA with different parameter efficiency tradeoffs.
Utilities
Image Generation
Generate images from the command line using trained models and LoRAs.
Model Tools
Merge models, resize LoRAs, convert between formats, and cache latents.