Google’s new AI tool Whisk uses images as prompts

Google has but another AI tool so as to add to the pile. Whisk is a Google Labs picture generator that allows you to use an present picture as your immediate. However its output solely captures your starter picture’s “essence” somewhat than recreating it with new particulars. So, it’s higher for brainstorming and rapid-fire visualizations than edits of the supply picture.

The corporate describes Whisk as “a brand new kind of inventive instrument.” The enter display begins with a bare-bones interface with inputs for fashion and topic. This easy introductory interface solely helps you to select from three predefined kinds: sticker, enamel pin and plushie. I think Google discovered these three allowed for the type of rough-outline outputs the experimental instrument is most perfect for in its present type.

As you possibly can see within the picture above, it produced a stable picture of a Wilford Brimley plushie. (Google’s phrases forbid photos of celebrities, however Wilford slipped via the gates, Quaker Oats in tow, with out alerting the guards.)

Whisk additionally features a extra superior editor (discovered by clicking “Begin from scratch” from the principle display). On this mode, you should utilize textual content or a supply picture in three classes: topic, scene and elegance. There’s additionally an enter bar so as to add extra textual content for ending touches. Nonetheless, in its present type, the superior controls didn’t produce outcomes that seemed something like my queries.

For instance, take a look at my try to generate the late Mr. Brimley in a lightbox scene within the fashion of a walrus plushie picture I discovered on-line:

Whisk spit out what seems like a vaguely Wilford Brimley-esque actor consuming oatmeal inside a lightbox body. So far as I can inform, that dude isn’t a plushie. So, it’s clear why Google recommends utilizing the instrument extra for “speedy visible exploration” and fewer for production-ready content material.

Google acknowledges that Whisk will solely draw from “a number of key traits” of your supply picture. “For instance, the generated topic might need a special peak, weight, coiffure or pores and skin tone,” the corporate warns.

To grasp why, look no additional than Google’s description of how Whisk works beneath the hood. It makes use of the Gemini language model to write down an in depth caption of the supply picture you add. It then feeds that description into the Imagen 3 image generator. So, the result’s a picture based mostly on Gemini’s phrases about your picture — not the supply picture itself.

Whisk is barely out there within the US, at the very least for now. You’ll be able to strive it on the mission’s Google Labs site.