A drawing of a horse and the text prompt "galloping" improved the drawing and set the animal in motion. (all images courtesy Stability AI, screenshots Elaine Velie/Hyperallergic)

Artificial intelligence (AI) image generators have been quite busy lately, from producing spooky art, strange videos, and a terrifying creation called “Loab” to helping a Colorado artist win a blue ribbon at the state fair this summer.

Generators like Midjourney and DALL-E normally use text prompts — along with their vast encyclopedias of imagery from the Internet — to formulate results. Now, Stable Diffusion’s AI allows users to create images from other pictures, drawn on the spot. That’s right: Input your napkin doodles and half-sketched masterpieces and AI will produce a complete image. (Well, sort of.)

Stable Diffusion is an open-AI platform, meaning that users can manipulate and change the code to achieve their desired results. One of those projects, called “Diffuse the Rest,” adds action, detail, and stylization to drawings and uploaded pictures. The tool is based on the work of Suraj Patil and comprises a set of collaboratively-created GitHub code. (Stable Diffusion’s own beta program, called “Dreamstudio,” also includes an image-to-image generator, but there is no drawing option.)

Users can draw or upload an image and then add a text prompt. In one example, Stable Diffusion turned a bad drawing of a horse into a better one, and following the text prompt “galloping,” placed the new rendering in motion.

Stable Infusion added the Loch Ness monster to a regular image of the lake. (original photo by Jon Roberts via Flickr)

We also fed Diffuse the Rest an image of a stormy Loch Ness Lake in Scotland, and following the simple text prompt “Loch Ness monster,” the generator added the storied creature to the photograph. Pairing the prompt “style of Picasso” with the Mona Lisa also produced realistic results.

The prompt “style of Picasso” with a photograph of the Mona Lisa

In a video, TikTok user Eric Hu explains the new feature and its potential application for design. He shows a sketch of a dog transformed into an impressive illustration, then a plug-in that turns a sketch of sneakers into a 3D-modeled illustration that looks ready to shop. Created by Antonio Cao, who posted the sneaker demo on Twitter, the plug-in is for design software Figma and utilizes the Stable Diffusion AI.

Using the Diffuse the Rest generator, a photograph of a sailboat coupled with the prompt “whales jumping” created a very strange image of whales flying through the sky like seagulls. Just like DALL-E, Stable Diffusion isn’t perfect, but it’s a fascinating example of the way in which AI is evolving to create increasingly sophisticated — and increasingly bizarre — content.

The generator couldn’t create a perfect image of “whale jumping.”

Elaine Velie is a writer from New Hampshire living in Brooklyn. She studied Art History and Russian at Middlebury College and is interested in art's role in history, culture, and politics.