What you need to know
- Google launched its latest AI experiment, Whisk, for Labs testers, which is aimed at giving users a quicker avenue to image generation.
- Whisk leverages Gemini and Imagen 3 and requires users to upload images to fit three areas: subject, scene, and style.
- Elsewhere, Google highlighted new version of Imagen 3 and Veo 2, which are rolling out today (Dec 16).
Google starts the week with a new AI experiment in Labs that U.S. users can start signing up for.
As detailed in a Keyword post, the newest test for Labs is called "Whisk" and goes about image generation in a new way: by letting users "prompt with images." This is an alternative to writing a detailed description, in case the words aren't flowing. Google explains that its newest model, Whisk, which leverages Gemini and Imagen 3, "extracts only a few key characteristics from your image."
To get started, the company says users must provide three images for the following areas: subject, scene, and style.
The AI model is said to consider everything to generate a new "unique" image, like a Whimsical Walrus or an image of a pink donut enamel pin. However, Google warns that its AI only considers things to capture the "essence" of whatever images the user has uploaded to Whisk.
Google says that this design is intentional as Whisk will not replicate whatever you've uploaded to the letter. The post adds that certain aspects of your images, such as subject height, hairstyle, and skin tone might vary, too. It understands that these attributes are important to users, which is why Google says Whisk gives users a "review and edit" feature after generation.
The company adds that this AI model is intended to give users a quicker experience with image generation than the typical methods. Moreover, Whisk lets users "remix" the images they've added to the creation tool for another round of ideas. After an image is created — using Gemini's model — Whisk will create a "detailed" description of what it created.
Labs testers in the U.S. can sign up today (Dec 16) for Whisk and try it out.
New Veo and Imagen Versions
In a separate blog post, Google detailed new versions of Veo 2 and Imagen 3. Beginning with the latter, the post says Imagen 3 has been updated to produce "brighter" and "better composed" images. Imagen 3 can reportedly follow your descriptions (prompts) better than before with richer details in its final renders.
The newest version of Imagen 3 is rolling out today (Dec 16) globally in ImageFX in Google Labs.
The latest updates to Imagen 3 arrive just before the year's end and several months after its (quiet) U.S. launch. Google started making its AI image generator available to users more prominently in August on its Vertex AI platforms. The company touts its safety precautions in place to avoid the creation of illegal or otherwise offensive imagery through Imagen 3.
Elsewhere, Veo 2 is receiving an update that widens the AI video generator's horizons. Google states this new version of Veo 2 lets users create "incredibly high-quality" videos for a "wide range of subjects and styles." Veo 2 can reportedly reach resolutions heights of 4K and it can supposedly understand more detailed descriptions of what you want.
The company gives the example of a user telling Veo 2 to utilize an "18mm lens" and states the AI will understand how to replicate a wide angled shot. Moreover, its understanding of expressions, movements, and more help to facilitate a more engrossed AI-generated video experience.
Google also states that Veo 2's latest updates help it hallucinate details "less frequently."
Today (Dec 16), Veo 2's update is rolling out in Google Labs' VideoFX and the company is slowly expanding the number of test participants. If you're interested, you can signup on the waitlist. Google then teased its plans to bring Veo 2 to YouTube Shorts and "other products" next year. Plans for YouTube incorporation are something we heard about earlier this year.
