Stable Diffusion VAE Optimization

6 repos

Fine-tuned and optimized Variational Autoencoder (VAE) models for Stable Diffusion and SDXL text-to-image generation pipelines. These repositories focus on improving VAE performance through different training objectives (MSE, EMA) and precision optimizations (fp16), addressing quality and efficiency tradeoffs in diffusion model inference. Repositories here provide drop-in replacements and fine-tuned weights compatible with the Hugging Face Diffusers library ecosystem.

Related clusters