Google Whisk: Use images as a prompt
You can enter images as prompts in the Google Whisk image generator instead of using words. A playful AI whisk.
Whisk turns three pictures into one.
(Image: Google Blogbeitrag)
Whisk is a new image generator from Google. It creates images by entering other images as a prompt. There is no room for words, at least in the first step. However, Whisk is initially only available in the USA. As it is an experimental image generator, it can only be accessed in the USA via Google Labs, Google's test environment.
Videos by heise
The term Whisk means whisk in German. So the image generator is supposed to mix images and perhaps even whisk them? If you look at the sample images from Google on the Whisk website, they are more reminiscent of gimmicks. Everything is comic-like. There are no photorealistic images. Google also writes that it is an experiment "that lets you use images as prompts in a quick and fun creative process." And: "Prompt less, Play more", i.e. "less prompting, more playing".
Gemini and Imagen work together
There is not one image as a prompt, but several images that can be used as a template for the different areas. One image should contain the subject, another represents the scenery and a third image is responsible for the style. The blog post says: "Then you can recombine them to create something unique, from a digital plush toy to an enamel pin or sticker."
Whisk is based on Google's AI model Gemini and the image generator Imagen 3. Gemini first describes the prompted images in the background, the texts then flow to Imagen 3 and are processed there. The underlying prompts, i.e. what Gemini describes, can be viewed and edited. Google says that it is less about pixel-perfect images and more about getting ideas off the ground quickly.
(emw)
