Google Labs has launched Whisk, a novel generative AI tool that lets users create images by prompting with other images, rather than just text. This new approach allows users to remix subjects, scenes, and styles in unique ways, pushing the boundaries of creative image generation.
Whisk’s workflow is intuitive: users drag and drop three images – one for the subject, one for the scene, and one for the style. The underlying Gemini model automatically generates detailed captions of these images, which are then fed into Google’s latest image generation model, Imagen 3. This process extracts the essence of the input images, rather than attempting to create an exact replica, resulting in novel and often surprising results.
Whisk emphasizes rapid visual exploration and creative experimentation. It’s designed to allow users to quickly generate multiple options, refining their ideas along the way, rather than focusing on precise pixel-perfect edits. Users can tweak results by adjusting prompts, exploring many different combinations of subject, scene and style.
Early testing with artists and creatives has shown that Whisk is seen as a new type of creative tool, fostering exploration and generating a range of diverse visuals. Examples shown include generating a fantastical fish with a city on its back, a whimsical walrus in a flower crown, an enamel pin of a sprinkled doughnut, and a sparkly horned cat on a lily pad. Users can easily download their creations to use in their creative projects.
Whisk’s launch is currently limited to the US. Google is emphasizing that this is an experiment and feedback from users is invaluable for improving the tool. It also highlights that because the tool extracts key characteristics, images might differ from expectations, such as subtle changes in height, weight or skin tone of the subject; which the user can then edit. The new tool is currently accessible at labs.google/whisk and Google encourages feedback via newsletter, Discord, Reddit and X.
Leave a Reply