The archive · Product Ideas · Product decision · 2024–2025
Google Whisk lets photos be the prompt — remix subject, scene, style (2024)
Google Labs' Whisk generates images by dragging in pictures for subject, scene and style — Gemini writes captions, Imagen 3 remixes them into new art.
Google (Google Labs)
What it had to solve
By late 2024, Google's image models still required long, carefully worded text prompts, which the Labs team felt slowed creative people down. Whisk's product managers said the experiment grew out of conversations with filmmakers, advertisers and fashion designers who wanted speed and play over pixel-perfect control.
How it works
Whisk launched in the US on 16 December 2024 as Google Labs' newest generative-AI experiment. Instead of typing a long description, users drag in images for three roles: the subject, the scene and the style. Users can supply several images per slot, add optional text, or press a dice button to let Google fill in suggestions.
Behind the scenes the Gemini model writes a detailed caption for each input image, and those captions go to Google's Imagen 3 model, which composes a new image. Because Whisk extracts only key characteristics rather than an exact replica, results can differ from expectations — so Google lets users view and edit the underlying prompts at any time. The team said the tool was built from conversations with filmmakers, advertisers and fashion designers, and described it as 'a new type of creative tool' for rapid visual exploration.
Reviewers found the loop genuinely playful: images generated in a few seconds, looked a little strange, and invited another iteration. In February 2025 Google opened Whisk to more than 100 countries, and walkthroughs showed how the preset styles (sticker, pin badge, capsule toy, lunch box and more) turned the subject and scene inputs into shareable keepsake-style art, cementing the tool as the showcase for prompting with images instead of words.
Why it lands
- It removes the wordsmithing barrier: uploading reference images is faster and more expressive than translating a visual idea into a long text prompt.
- Separating subject, scene and style turns generation into a remix kit — swap any one ingredient and the whole image changes.
- Because Gemini captioning captures only an 'essence,' results surprise rather than copy, which fits exploration and play.
- Showing the generated caption as an editable prompt turns every near-miss into a next step instead of a dead end.
What it did
Google framed Whisk as 'rapid visual exploration, not pixel-perfect edits' and 'a new type of creative tool' rather than a traditional image editor; The Verge's hands-on found it 'fun to iterate on' even when results were strange. Two months after the US launch, Google expanded Whisk to more than 100 countries, making image-first prompting the defining showcase of its Labs experiments.
What you can take
Match the tool to the user's native material: creators think in reference images, so Whisk made pictures the prompt — and surfaced the AI caption as an editable handle for control.
Since then
Whisk stayed a Labs experiment rather than a flagship product, but it traveled fast: from a US-only launch on 16 December 2024, Google expanded it to more than 100 countries in February 2025, including Japan and Brazil. Google's vice president Josh Woodward described it at launch as built on conversations with filmmakers and creatives, and Google Labs continued to iterate on it as part of its FX toolset. The idea Whisk normalized — using pictures, not paragraphs, as the prompt for AI image generation — became a visible part of Google's generative-AI story.
Sources
- Whisk: Visualize and remix ideas using images and AI
- Google's Whisk AI generator will 'remix' the pictures you plug in
- Google's image generation AI 'Whisk' that mixes multiple images is now available in over 100 countries
spotted an error? The archive wants to know.
Your turn
You just read one. Describe the brief you are staring at, and see who has been given the same problem.
Free account · 3 free questions · no card