Stable Diffusion.cpp is an open-source implementation of Stable Diffusion models optimized for CPU and GPU inference. Written in pure C/C++, it offers a lightweight and efficient alternative to traditional frameworks, supporting models like SD1.x, SD2.x, SDXL, and SD3/3.5. The project emphasizes minimal dependencies, low memory usage, and broad platform compatibility, including Windows, macOS, Linux, and Android (via Termux).
Key Features
Model Support: Compatible with SD1.x, SD2.x, SDXL, and SD3/3.5 models.
Quantization: Supports 2-bit, 3-bit, 4-bit, 5-bit, and 8-bit integer quantization for reduced memory usage.
Precision: Offers 16-bit and 32-bit floating-point precision.
Sampling Methods: Includes Euler A, Euler, Heun, DPM2, DPM++ 2M, DPM++ 2M v2, DPM++ 2S a, and LCM.
ControlNet Support: Integrates with ControlNet for enhanced image generation.
LoRA Support: Compatible with LoRA models for fine-tuning.
VAE Support: Handles standalone VAE models without conversion.
Flash Attention: Utilizes Flash Attention for memory optimization.
Cross-Platform Compatibility: Runs on Windows, macOS, Linux, and Android (via Termux).
Low Memory Usage: Requires approximately 2.3GB of RAM for generating a 512x512 image with fp16 precision.
Android DevHub
.png)
.png)
.png)