Universal Checkpoint Blender (SDXL & SD 1.5)
A lightweight, standalone Python tool for blending Stable Diffusion checkpoints and "baking" LoRAs directly into model weights. This tool was designed as a RAM-efficient alternative to bloated WebUIs, allowing for precise control over model architecture without the overhead.
β¨ Key Features
- Architecture Agnostic: Works with SDXL, Pony, Illustrious, NoobAI, and SD 1.5 models.
- Smart Blending: Adjust the ratio between Model A (Base) and Model B (Mix-in) with a precise slider.
- LoRA Injection: "Bake" a LoRA directly into the resulting checkpoint with adjustable strength (0.1 - 2.0).
- RAM Optimized: Includes an aggressive memory flush system to allow multiple merges in one session on 32GB (or more) systems.
- Space Saving: Optional "Prune to FP16" checkbox to keep your model files at standard sizes (~6GB for SDXL, ~2GB for SD 1.5).
- Simple UI: The "Universal Model + LoRA" interface is designed for anyoneβjust browse, slide, and blend.
π How to Use
1. Install Requirements
Ensure you have Python installed, then install the necessary "math engines" via terminal: pip install torch safetensors
2. Run the App
Download the script and run it from your terminal or command prompt: python blender.py
3. Blending Logic
- Model A: Your primary base model.
- Model B: The model you want to "flavor" the base with.
- The Slider: Moving the slider to 0.15 adds 15% of Model B into Model A. This is the "sweet spot" for adding Pony/Illustrious logic to standard SDXL models without breaking prompt adherence.
π Technical Notes
- LoRA Baking: This is a "destructive" merge that permanently fuses the LoRA into the weights, making it portable for any UI that supports standard .safetensors files.
- History: The tool automatically creates a merge_history.txt file in its folder, acting as a "recipe book" for every model you create.
- OS Support: Tested and confirmed on Ubuntu 20.04+, but works on Windows 10/11 and macOS as well.
π€ Credits
Designed and built by Human and A.I.
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. π Ask for provider support