--- tags: - stable-diffusion-xl - text-to-image - gguf - quantization - unet - vae - clip license: mit base_model: stabilityai/stable-diffusion-xl-base-1.0 datasets: - stabilityai/stable-diffusion-xl-base-1.0 - stabilityai/stable-diffusion-xl-refiner-1.0 model_creator: stabilityai model_type: stable-diffusion-xl task: text-to-image timestamp: 2025-03-06 --- # SDXL GGUF Quantized Model This repository contains a quantized version of **Stable Diffusion XL** in the **GGUF** format. The model has been converted to different quantization levels, including **Q4_K_S, Q5_K_S, and Q8**, allowing for flexible deployment based on hardware capabilities. The UNet, VAE, and CLIP components are provided separately for better optimization and compatibility. ## Quantization Details | Component | Available Quantization | |-----------|----------------------| | UNet | Q4_K_S, Q5_K_S, Q8 | | VAE | FP16 | | CLIP | FP16 | ## Files & Structure - `sdxl-unet-q4_ks.gguf` - `sdxl-unet-q5_ks.gguf` - `sdxl-unet-q8.gguf` - `sdxl-vae-fp16.safetensors` - `sdxl-clip-fp16.safetensors` Each quantization level offers a trade-off between speed and quality. **Q4_K_S** provides the highest speed but lower quality, while **Q8** retains more details with higher VRAM usage. ## Usage This model can be used with any **GGUF-compatible** inference engine, such as **ComfyUI**, **Kohya's SDXL GGUF loader**, or **custom scripts supporting GGUF-based SDXL inference**. ## Hardware Requirements - **Q4_K_S**: Suitable for low-VRAM environments (2GB+) - **Q5_K_S**: Balanced performance and quality (3GB+ VRAM recommended) - **Q8**: Best quality, requires higher VRAM (4GB+ recommended) ## Acknowledgments This model is based on **Stable Diffusion XL** by [Stability AI](https://stability.ai/) and has been quantized for improved accessibility across various hardware configurations. For support and discussions, feel free to open an issue or reach out on Hugging Face forums! ---