siliconflow/onediff

Health v2 · maintenance only

Adoption not part of the score

1964 stars · 128 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

OneDiff is an out-of-the-box acceleration library for diffusion models, providing PyTorch compilation tools and optimized GPU kernels. It integrates with popular frameworks and UIs such as Hugging Face diffusers and ComfyUI to speed up image and video generation.

Use cases

• speed up stable diffusion image generation
• accelerate sdxl inference in comfyui
• compile pytorch diffusion models for faster gpu inference
• run flux or sd3 models faster
• quantize diffusion models to reduce memory usage
• accelerate stable video diffusion
• serve stable diffusion models with lower latency

When to choose

• you use ComfyUI or HF diffusers and want faster diffusion inference
• you need GPU kernel optimization and compilation for SDXL, FLUX, or DiT models
• you want quantization to cut memory and boost throughput

When to avoid

• you need the latest diffusers versions beyond the supported maximums
• you work on CPU-only or non-CUDA hardware
• you need a general-purpose inference engine beyond diffusion models

Facets

library · maturity active

machine-learning llm-inference gpu-computing compiler deep-learning image-processing gpu-computing machine-learning python windows diffusion-models stable-diffusion sdxl comfyui diffusers inference-acceleration pytorch lora quantization aigc gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="siliconflow/onediff")