Member-only story
Stable Diffusion vs Disco Diffusion
It’s time to ART!
With the beta release of Stable Diffusion on Monday, we are entering a new era of AI-generated images and content, together with many great products on the market, Dalle-2, Midjourney, Disco Diffusion, and Imagen.
We will compare some generated images and videos from both models and have a quick summary at the end. If you want to have a deep dive into generative AI models, especially what is diffusion mode, highly recommend checking out Lilian’s blog here.
Stable Diffusion
An open sourced text to image model from Stability AI. Like Dalle-2, it can generate images based on text prompt in seconds.
One of the breakthroughs it has is a good balance between quality and generation speed, and they can even run on a single GPU with less than 16G VRAM with reasonable quality.
From their release notes, the model is potentially working with AMD and Apple M1/M2 chipsets in the future release. Looking forward to having this “billion dollar model” runs on your computer for numerous applications.
Pre-trained models can be downloaded here: https://huggingface.co/CompVis/stable-diffusion
Try our their DreamStudio for rapid image generation: http://beta.dreamstudio.ai
Disco Diffusion
Disco Diffusion (aka DD) is a clip-guided diffusion model that can generate amazing images from text prompt, and it is pretty impressive in generating abstract art, with vivid color combination, and sometimes mind-blowing image composition and details.
It starts from Katherine Crowson‘s notebook and her fined tuned diffusion model, then evolved and optimized by many other developers, together with projects like CLIP from Openai and Openclip from the community.
Image Comparison
In the following comparison, the same prompt has been used to feed into each model with default parameters to generate the image. Some prompts are borrowed from Ethan Smith‘s famous traveler guide to latent space.
Prompt: “a boy looks outside his bedroom window to see the beautiful cosmos, trending on artstation”
