Google Whisk: Mix Images with AI — New Creative Tool

How Whisk Works

Google Whisk, released January 27, 2026 via Google Labs, introduced a novel three-input image generation paradigm. Rather than generating from a text prompt alone, Whisk accepted three optional image references: a Subject (what appears in the image), a Scene (the environment or setting), and a Style (the visual aesthetic). Imagen 3 then synthesised a new image combining the subject from the first image in the setting of the second, rendered in the style of the third.

For example: input a photo of your product as Subject, a luxury hotel lobby as Scene, and a high-fashion photography portfolio as Style, and Whisk generated a series of professional product-in-context images without a photographer, a model, or a location booking.

Imagen 3 Under the Hood

Whisk was powered by Imagen 3, Google's most capable text-to-image model, trained on a licensed image dataset. Unlike text-prompt-only generation, Whisk extracted semantic understanding from the reference images using Google's visual embedding models, translating image content into prompt conditions for Imagen 3 rather than requiring users to describe what they wanted in words.

Remix and Iteration

Generated images could be remixed with a single click, generating variations that maintained the same semantic combination with different visual interpretations. Users could also adjust the influence of each reference image — increasing subject fidelity for product images, increasing style influence for more stylised artistic outputs, or emphasising the scene for environmental shots.

Availability and Limits

Whisk launched as a free Google Labs experiment accessible via labs.google.com/whisk with a Google account. Free usage was rate-limited but generous — 10 generations per session, refreshing daily. The quality of outputs was consistently strong across categories, with particular excellence in fashion, product, and architectural imagery.

What This Means for Indian Businesses