Google Whisk
A recent project from Google Labs called Whisk allows you to prompt with visuals for a quick and enjoyable creative process.
Google is introducing Whisk, its newest generative AI experiment, in the US today. With Whisk, you may prompt using images instead of long, detailed text prompts. Just drag and drop pictures to begin designing.
Google Whisk allows you to add three images: one for the scene, one for the topic, and one for the style. After that, you can remix them to make something entirely original, like an enamel pin, sticker, or digital plush toy.
The Gemini model automatically creates a thorough explanation for your photos in the background. After that, it feeds those descriptions into Imagen 3, Google’s most recent picture generating model. This method captures the spirit of your subject, not a perfect duplicate. In this manner, you can quickly and creatively modify your subjects, sceneries, and styles.
Whisk may produce photos that are different from what you had in mind because it simply pulls a few essential elements from your image. For instance, the created subject may differ in skin tone, hairdo, height, and weight. It allows you to inspect and modify the underlying prompts at any time since it recognizes that these features might be essential to your project and that Whisk might fall short.
People have described Google Whisk as a new kind of creative tool rather than a conventional image editor during its early testing with artists and creatives. Google didn’t build it for pixel-perfect tweaks but for quick visual exploration. It lets you work through dozens of alternatives and download the ones you like most, all while allowing you to explore ideas in fresh and imaginative ways.
Whisk adds Google Launchpad to recipe for success
Foodies can receive personalised recipe ideas from Google Whisk based on their purchasing habits, saved recipes in their cookbook, and other factors. The service is used globally and is integrated into numerous prominent food publisher websites. Whisk got involved with Google Launchpad because he wanted to connect with its users. Organising testing and interviews took weeks, despite the team’s desire to include more user input in the design process. After learning about it from other start-ups, they reached out to Google and participated in a Design Sprint session to enhance their user experience.
Play more, prompt less!
We think that creating photographs shouldn’t require you to “learn how to prompt.” Visual experimentation, iteration, refinement, and remixing of concepts should be simple. similar to how you would with a friend. Thus, Google attempting something different!
The newest generative imagery experiment from Google/FX focusses on quick visual inspiration without requiring a thorough understanding of prompts!
Simply provide a few photographs for general direction (location, subjects, styles), and Google Whisk will attempt to distil the essence of each to recommend a few images for you to continue brainstorming.
The Gemini model automatically creates a thorough explanation for your photos in the background. After that, it feeds those descriptions into Imagen 3, Google’s most recent picture generating model.
Whether it’s making a fantastic holiday card, converting a painting into a plush toy, or imagining the start of a narrative… We can’t wait to follow your journey with Google Whisk.
Using Whisk to create is easy!
Get ready
Bring in visual components so Google Whisk may examine and integrate them. You can upload an image from a folder or drag & drop it. Additionally, you can make a basic reference using a text prompt, alternatively use the “roll the dice” or “inspire me” options to help us generate a few ideas.
Behind the scenes: Gemini uses its visual comprehension to title these items. Google Whisk makes use of these text descriptions. To check if we got it right and make any necessary adjustments, click edit!
Examine
It’s time to get things going! One or more subjects, one scene, and one style are among the assets you can choose and use. Those will be combined by the system to create imaginative remixes.
Keep riffing and see what Google Whisk thinks of! In order to keep your imagination active and play with details, you might also include some light instruction.
“Instruct the characters to consume ice cream.”
“The cat and dinosaur are giving each other high fives!”
“Verify that the enamel pin is circular.”
“Change the colour palette to a pastel one.”
Behind the scenes: Gemini creates the prompt for you by using the captions and your instructions. To view what it has been whispering to Imagen 3, click Edit.
Make it better
See a picture you like, but perhaps the hat ought to be blue? Or does it lack a background sunset? To request tiny to medium adjustments that remain directionally near to the original, enter refine mode.
Behind the scenes: Gemini uses your input to refresh the prompt! We urge the model to remain near, but we still regenerate every pixel from that prompt.
Diagnose
Let’s face it, things could take a crazy turn! Perhaps some components were omitted? Perhaps the precise item you’re searching for simply doesn’t fit?
By clicking the prompt button or icon at any of the aforementioned stages, you can identify the underlying prompts, make changes, manually enter those important data, and request that the model produce additional possibilities. You are ultimately in charge.
What are the meanings of the categories?
Subject
That’s what the picture depicts! Characters, items, or a mix of both. A vintage rotary telephone! An awesome chair! A cinema screen made of cardboard. An enigmatic vampire from the Renaissance. Additionally, you might use yourself as a guide and see the results.
Scene
Where will the subjects appear? A runway for fashion? A festive card that pops up? Characters can be added to the scene alongside those who are already present, or perhaps they can be switched out. Worth a shot.
Style
Perhaps you would like to provide more details about the style, substance, or method used to depict the above. That’s what style is for. To emphasise that advice, feel free to indicate your top priorities in the main prompt box.
When you include additional details (such as “our subjects having a birthday dinner”), you can use natural language to refer to them, and Whisk will attempt to include that.
Tutorials
To help you understand how this functions naturally in the tool, Google supplied a few methods.
Playground: Google’s landing page offers a condensed version of the tool so you may experience the magic with just one click. Enter a picture and watch it become a cuddly toy! (or enamel pin! or sticker!)
When you select “start from scratch,” this button will appear. Certain assets will be pre-populated. offer advice and walk you through the main user interface’s main sections so you can produce your initial results. Simple!
To get started, quickly add a few subject, scenario, and style suggestions using the dice roll, which is situated at the top of the left panel. or continue to riff!
How does Whisk operate?
We must first have a comprehension of each image you cite before we may combine aspects from various photographs. Gemini’s multimodal knowledge is useful in this situation! Gemini is used by Google Whisk to visually comprehend the photographs you upload and produce text descriptions (also known as captions) about them. Translate the image to text (I2T), to put it another way. The purpose of these descriptions is to help you remix ideas by capturing the essence of your references rather than reproducing the original.
These captions are then used to create a comprehensive prompt that uses Google’s newest and most potent picture-generating model, Imagen 3, to create an image according to your instructions. Alternatively, T2I stands for text-to-image translation.
The aforementioned procedure aids Google Whisk in better comprehending and expressing the concepts you’re developing, and iterates during the conversation.
How can I change that? It’s not my character
This is deliberate. Only a few essential features are taken from the picture you supply to help the model in our experiment. Google’s objective is to capture the spirit of the subject, not to produce an identical reproduction.
As a result, the resulting image can look different. For instance, the created subject may have a different skin tone, hairdo, height, or weight. It recognizes that these characteristics might be essential to your character’s distinct individuality. Google invites you to tweak your directions and provide more specific recommendations to get a result that is closer to your vision.
You can provide with feedback using the option on the upper right.
Where can I find Whisk?
Google Whisk uses English text inputs and is only available in the United States at this time. Soon, it hopes to extend to further nations!
Can I show off what I made?
Yes, to save and share, simply click the download symbol.
