Whisk

Overview

Whisk’s a free test tool you just open it in a browser and mess around with it.

Whisk came from Google Labs with help from DeepMind. It’s built to let people think with images, not words. Powered by Google's Imagen 4 image model.

You can fine-tune your result tweak the prompts, save the image or download it. Want to make more versions of the same idea? No problem, Whisk handles that too.

Originally just for the US it now works in 100+ countries. Some places like the EU or UK might not have access yet though. Whisk just hit 77 more countries. Now it also works with Veo 3. You can turn Whisk-made images into 8-second clips with better detail, real-looking scenes and sound. Everyone gets 5 free clips a month. If you're on Google AI, you can make more to support your work.

You don’t need to write long prompts. Instead Whisk lets you upload up to 3 pics. One sets the subject one sets the scene and one sets the style. If you don’t got all three no worries Whisk will fill in with random ones.

Once the pics are up Whisk uses Gemini to read them and write captions from what it sees. Then Imagen 3 steps in and makes new visuals out of that. The goal ain’t to copy exactly but to remix things. It feels natural fast and playful.

Whisk now has Precise Mode with Veo 3. 

Precise Mode works different from the usual “Creative mode.” That older mode sends your image to Gemini to get captioned before it hits the model. The new one skips that and uses your uploaded image right away. It runs on a custom Imagen 4 setup that sticks to faces, style, and scene better. Then Gemini 2.5 Flash steps in to fix up the final version.

Here’s how to get going.
Turn on Precise ref. Flip the toggle in the prompt bar settings.
Pick your references. Add up to 3 images for subject and scene, and 1 for style. Use a clear front-facing color photo, not blurry, angled, or filtered ones, and don’t include other people. No photorealistic kids.
Choose at least 1. You need either a scene or subject image, or one style image.
Add a short prompt. It helps guide the tool.