Stable Diffusion 2.0 has just been released, and it again shows how fast this technology is evolving and improving, especially when it’s open-sourced.
You can read the complete announcement here, but here’s the tldr:
- more powerful, but still runs on a single GPU (to make it accessible for as many people as possible)
- greatly improved text-to-image diffusion models
- images with default resolution of up to 768×768 pixels
- more powerful upscaling (now 4x upscaling is possible, so a low-res 128×128 image can be upscaled to 512×512, and 512×512 images can be upscaled to 2048×2048 pixels)
- new Depth-to-Image Diffusion Model
- better inpainting
Want to play around with Stable Diffusion 2.0? It’ll be available in a couple of days on https://platform.stability.ai/ and DreamStudio
Leave a Reply