DiffSynth-Studio
The framework behind Qwen-Image-i2L — the first "Image→LoRA" visual generation system
The framework behind Qwen-Image-i2L — the first "Image→LoRA" visual generation system: one image generates a LoRA style module.
Image generation and computer vision — detection, editing, 3D.
Key features
- Open-source Diffusion model engine supporting image, video, and audio generation models
- VRAM management scheduling parameters across disk, memory, and VRAM for consumer GPUs
- Parameter quantization (NF4, INT8) reducing VRAM needs for inference and LoRA training
- Almost every inference-supported model also supports training (base, LoRA, adapters)
- ComfyUI node package and WebUI LoRA training tool built on top
Good for
- Running and training diffusion models on consumer-grade GPUs
- Developers integrating diffusion pipelines into ComfyUI workflows
Stack & tags
PythonPyTorchdiffusionenginevideo-generationimage-generationlora-trainingquantizationmodelscope
Quick start
pip install DiffSynthFrom the project README on GitHub.
Similar open-source projects
FLUX.2 klein
★ 25,996Black Forest Labs' sub-second image generation — generate and edit in <0.5s
TRELLIS.2
★ 12,083Native and Compact Structured Latents for 3D Generation