Stable Diffusion logo

Stable Diffusion

Free

Stable Diffusion by Stability AI is a state of the art text-to-image model that generates images from text. #opensource

FreeFree tier
Inputs: textOutputs: image
Type
Open Source

About Stable Diffusion

Stable Diffusion is a latent text-to-image diffusion model capable of generating photo-realistic images given any text input. It uses a fixed, pretrained CLIP ViT-L/14 text encoder and was fine-tuned on the laion-aesthetics v2 5+ dataset at 512x512 resolution. The model is open source under the CreativeML OpenRAIL M license and can be run using the Hugging Face Diffusers library, supporting PyTorch and JAX/Flax frameworks. The v1-4 checkpoint was initialized from v1-2 and further fine-tuned for 225k steps with 10% text-conditioning dropout to improve classifier-free guidance sampling.

Key Features

Latent text-to-image diffusion model
Uses CLIP ViT-L/14 text encoder
Fine-tuned at 512x512 resolution on laion-aesthetics v2 5+
Classifier-free guidance sampling with 10% conditioning dropout
Compatible with Hugging Face Diffusers library
Supports multiple schedulers (PNDM, Euler, etc.)
Available in PyTorch and JAX/Flax
Optimized for <4GB GPU VRAM with float16 and attention slicing

Pros & Cons

Pros
  • Open source and freely available under CreativeML OpenRAIL M license
  • State-of-the-art image quality with latent diffusion
  • Easy integration with Hugging Face Diffusers and other frameworks
  • Active community support and extensive documentation
  • Optimizations available for less powerful GPUs
Cons
  • Requires a GPU with sufficient VRAM (optimizations enable <4GB usage)
  • Inference can be slow without hardware acceleration
  • May produce artifacts or reflect biases in training data
  • License imposes restrictions on commercial use

Best For

Generate images from descriptive text promptsCreative design and art generationStoryboarding and concept visualization

FAQ

What is Stable Diffusion v1-4?
It is a latent text-to-image diffusion model fine-tuned from v1-2 on the laion-aesthetics v2 5+ dataset at 512x512 resolution, using a CLIP ViT-L/14 text encoder.
How can I run Stable Diffusion?
Use the Hugging Face Diffusers library: install diffusers, transformers, and scipy, then load the model with StableDiffusionPipeline and run it on a GPU with torch.float16 for memory efficiency.
What license does Stable Diffusion use?
The model is released under the CreativeML OpenRAIL M license, adapted from BigScience and the RAIL Initiative.
What hardware is recommended?
A GPU with at least 4GB VRAM is recommended. For less than 4GB, use float16 precision and enable attention slicing.