SD-Trainer v2.0.0

SD-Trainer v2.0.0

New Features

• Anima LoRA Training - Full support for LoRA / LoKr (LyCORIS) / T-LoRA
• T-LoRA - Timestep-Dependent LoRA with dynamic rank and orthogonal init
• Train Monitor - Interactive ECharts Loss chart with zoom/pan/restore, real-time progress
• Windows Portable Package - Lightweight ~15MB download, auto-installs PyTorch+CUDA on first launch
• Update Check - Non-blocking version check against GitHub Releases on startup
• Utility Scripts - Update-SD-Trainer.bat / Download-Anima-Model.bat included

Bug Fixes

• Fix bf16 training crash - Auto-enable full_bf16 when mixed_precision=bf16 to prevent dtype mismatch (BFloat16 != float)
• Fix submodule clone failure - vendor/sd-scripts is now directly included (no more --recurse-submodules needed)
• Fix install-cn.ps1 encoding - Proper UTF-8 BOM for Chinese characters on Windows PowerShell
• Fix commit check crash - verify_pinned_commit now warns instead of crashing on commit mismatch

Portable Package Usage

1. Download \SD-Trainer-v2.0.0.7z\ and extract
2. Double-click
   un_gui.bat\
3. First launch auto-installs PyTorch + dependencies (~3GB download, auto-detects CN mirrors)
4. Browser opens http://127.0.0.1:28000

Requirements: Windows 10/11 64-bit, NVIDIA GPU (RTX 20+), ~7GB disk space

Install from Source

\\�ash
git clone https://github.com/wochenlong/lora-scripts-next.git
cd lora-scripts-next

Windows: double-click run_gui.bat

Linux: bash install.bash && bash run_gui.sh

\\