4 things to know about Stable Diffusion 2.0

Stability AI just announced the latest version of its deep learning, text-to-image model

Image by Santiago Estela from Pixabay

A few hours ago, Stability AI announced the latest version of its popular deep learning, text-to-image model: “Stable Diffusion 2.0”.

Here are four things to know:

• The text-to-image models in this version can generate images with both 512x512 and 768x768 pixel resolutions by default.
• The release contains an Upscaler Diffusion model that improves the picture resolution by four. Stable Diffusion 2.0, when combined with text-to-image models, can now create pictures with resolutions of 2048x2048 or higher.
• A novel depth-guided stable diffusion model, depth2img, expands the V1 image-to-image capability with brand-new creative application possibilities. Depth2img infers the depth of an input picture (using an existing model) and then creates new images utilizing both text and depth data.
• The release also includes a new text-guided inpainting model, adjusted on the new Stable Diffusion 2.0 foundation text-to-image, which makes it simple to replace portions of a picture.

This should please generative AI enthusiasts around the world…

This story was first published on The PhilaVerse (my Substack newsletter).