Hugging Face released StableDiffusionPipeline in Diffusers alongside the public launch of Stable Diffusion v1.4, standardizing modular Python access to VAEs, U-Nets, and interchangeable noise schedulers.
Hugging Face Diffusers
Also known as: Diffusers · DiffusionPipeline
Plain English
The standard open-source Python library for loading, running, and fine-tuning image and video diffusion models—doing for visual generators what Hugging Face Transformers did for language models.
Technical Overview
Built on PyTorch by Hugging Face, diffusers decouples visual generative models into swappable components: ModelMixin backbones (U-Nets and DiTs), VAEs, text encoders, schedulers, and DiffusionPipeline orchestrators, with built-in Safetensors loading, LoRA/ControlNet injection, and CPU/GPU memory offloading.
Timeline (4)
2022
Release2023
ReleaseDiffusers
v0.14.0added first-class support for ControlNet conditioning and lightweight LoRA adapter training and merging, turning the library into the reference code backbone for academic and production diffusion fine-tuning.2024
ReleaseDiffusers
v0.30.0integrated FLUX.1 and CogVideoX, expanding the library's core abstractions from 2D convolutional U-Nets to 12B+ flow-matching Diffusion Transformers and 3D causal video autoencoders.2025
ReleaseDiffusers
v0.33.0expanded native Quantization backends (bitsandbytes, GGUF, TorchAO FP8) and unified open-weight video pipelines across Wan 2.1, HunyuanVideo, and LTX-Video.
See also (20)
Open full graphHugging Face Diffusers
15 more connections
- Sampling & SchedulersOutgoing · Implements
- CogVideo & CogVideoXOutgoing · MentionsIncoming · Part of
- ControlNet & IP-AdapterOutgoing · Mentions
- Diffusion Transformer (DiT)Outgoing · Mentions
- FLUXOutgoing · Mentions
- FP8Outgoing · Mentions
- GGUFOutgoing · Mentions
- HunyuanVideoOutgoing · Mentions
- Latent Space & VAEOutgoing · Mentions
- LoRA & QLoRAOutgoing · Mentions
- LTX-Video & LTX-2Outgoing · MentionsIncoming · Mentions
- QuantizationOutgoing · Mentions
- Stable DiffusionOutgoing · Mentions
- Video Generation Models (Sora, Veo, Kling, Wan)Outgoing · MentionsIncoming · Mentions
- WanOutgoing · Mentions