Google Labs unveils Whisk, an AI experiment that creates images using images

Google is releasing a new AI tool called Whisk, developed by Google Labs. It allows you to create images by inputting images rather than text. You can drag in photos for the subject, scene, and style, then remix them to create something unique. Whisk uses the Gemini model to write a detailed caption of your images automatically and then feeds those descriptions into Imagen 3 to generate the final image.

The tool is available on the web at labs.google/whisk, and Google is asking users to share their feedback. Google says that while Whisk doesn’t always create images that are true to the input, it’s meant to be a “new type of creative tool” for rapid visual exploration.

Google’s AI tools have been generating controversy of late. Earlier this year, Google unveiled Gemini, its “state-of-the-art” image model, and Imagen 3, which produces images from textual descriptions. However, Gemini and Imagen 3 have come under fire for generating images that reinforce harmful stereotypes.

In response, Google tweaked Gemini and Imagen 3 to reduce the likelihood that the models would generate images of, for example, people in front of fire trucks or doctors in lab coats. However, as Google has acknowledged, it’s challenging to eliminate the risk of models generating problematic images ultimately.

The company also recently released Veo, a video generation model, which has also been criticized for generating content that could be used in harmful ways.

In a blog post on Whisk, Google says that it understands “these features may be crucial for your project and Whisk may miss the mark,” but that Whisk is about “allowing you to work through dozens of options and download the ones you love.”

“Whisk is a new kind of visual exploration tool,” Thomas Iljic, director of product management at Google Labs, said in a statement. “In our early testing with artists and creatives, people have described Whisk as a new creative tool — not a traditional image editor. We built it for rapid visual exploration, not pixel-perfect edits.”

Iljic added that while Whisk is launching in the US today, it’s part of a more significant effort by Google to cook up experiments with generative AI models.