← Back to Blog

Google Whisk: Generate Stunning Images by Uploading Images (Not Just Text)

⏱ 19 min read • By

Google Whisk is a revolutionary free AI image generator that uses images as prompts instead of text descriptions. Upload a subject, a scene, and a style — and Whisk combines them into something entirely new. This complete guide covers every feature, workflow, and real-world use case for creators, marketers, and designers.

Google Whisk: Generate Stunning Images by Uploading Images (Not Just Text)

Every other AI image generator asks you the same question: describe what you want in words.

That works — until it does not. Until you are staring at a product photo that is exactly the right subject but in the wrong setting. Until you have a reference image that captures the aesthetic you want but you cannot articulate it in text. Until you want your brand mascot in a completely different context and every text prompt you write produces something slightly, frustratingly wrong.

Google Whisk solves this differently. Instead of making you translate vision into language, it lets you communicate the way creative professionals actually think — by showing reference images.

You provide three images: a Subject (what you want in the image), a Scene (where it should be), and a Style (how it should look). Whisk’s Imagen 3 model synthesizes these three visual inputs into a new image that honors all three references without copying any of them directly.

The result is something genuinely new — a tool that produces more predictable, more controllable creative outputs than text-only generation, particularly for users who have a clear visual reference but struggle with prompt engineering.

This is the complete guide to everything Whisk can do.

🔗 This is Post #13 in our Google AI series. Whisk builds on the foundation of Google Labs (Post #12), where it lives as an experiment. For AI image generation through text prompts, see Google AI Studio which provides API-level access to Imagen. For using generated images in presentations, see Google Slides AI.

What Is Google Whisk? The Core Concept

Whisk is built on Imagen 3, Google DeepMind’s highest-quality image generation model. What makes Whisk distinctive is not the model itself — it is the input mechanism.

Standard text-to-image generation workflow:

  1. Write a detailed text description
  2. Generate
  3. Iterate on description based on what came out wrong
  4. Repeat until acceptable

Whisk’s image-to-image-to-image workflow:

  1. Upload or describe a Subject image (what you want featured)
  2. Upload or describe a Scene image (the setting or environment)
  3. Upload or describe a Style image (the visual aesthetic)
  4. Generate — Whisk creates a synthesis of all three
  5. Iterate by swapping any individual input

The fundamental insight behind Whisk: a picture communicates more about what you want than a thousand words. Showing Whisk a photo of a mid-century modern chair as your Style reference tells it more about your aesthetic than any textual description could.

The Technical Reality: Remixing, Not Replicating

An important clarification: Whisk does not simply paste your subject into your scene with your style applied. It uses your inputs as semantic guidance — understanding the visual essence of each — and generates an entirely new image that synthesizes those elements.

This means:

  • Your subject will be recognizable but reimagined in the new context
  • Fine details from your style reference will inform the aesthetic without being directly copied
  • The output is Whisk’s interpretation, not a collage

This is both a strength and a limitation depending on what you need. For creative ideation and exploration, it is liberating. For precise brand reproduction, it requires more iteration.

How to Access Google Whisk

  1. Go to labs.google.com and find the Whisk experiment card
  2. Sign in with a Google account
  3. The Whisk interface loads — three input panels (Subject, Scene, Style) and a generate button
  4. No payment required for the free tier during Labs phase

Step 1: Understanding the Three Input Types

The Subject

The Subject is the main focus — the element that will be present in the generated image. Think of it as “what this image is about.”

What works well as a Subject input:

  • Product photos (clear, well-lit, with minimal complex background)
  • Character designs or illustrations
  • Animals or pets
  • Objects with distinctive silhouettes
  • Logos or mascots (with appropriate rights consideration)
  • Food items
  • Fashion items

Subject input tips:

  • Use images with a clear, uncluttered background — Whisk finds it easier to isolate the subject
  • High-contrast images give more recognizable results
  • Simple, iconic shapes transfer better than complex scenes
  • If you do not have a specific subject image, you can type a text description instead

What NOT to use as Subject:

  • Photos of real, identifiable people without their explicit consent
  • Trademarked characters, brand mascots you do not own (Disney, Marvel, etc.)
  • Images you do not have rights to use
  • Very low-resolution images (quality in, quality out)

The Scene

The Scene defines the environment, setting, or context in which the subject will appear.

What works well as a Scene input:

  • Environmental photos (forests, cities, kitchens, offices, beaches)
  • Architectural or interior spaces
  • Abstract backgrounds (gradient photos, texture references)
  • Seasonal or weather conditions (snow, golden hour light, rainy streets)
  • Conceptual settings (outer space, underwater, fantasy landscapes)

Scene input tips:

  • Scenes with strong, clear environmental identity transfer better
  • Lighting in the scene reference influences how the subject is lit in the output
  • Abstract or simple scenes give you more control; complex scenes may produce unexpected results
  • If you want the subject on a plain color background, type “clean white studio background” in the text field instead of uploading a scene image

The Style

The Style defines the visual aesthetic — the artistic treatment, photographic approach, or rendering technique.

What works well as a Style input:

  • Paintings by artists whose style you want to reference (with ethical consideration)
  • Photography style references (editorial, documentary, commercial)
  • Illustration styles (flat design, detailed line art, watercolor, oil painting)
  • Color palette reference images
  • Design system screenshots (for generating UI-adjacent content)
  • Specific era aesthetics (1970s advertising, Art Deco, Bauhaus)

Style input tips:

  • Style transfers most reliably with distinctive, unmistakable aesthetics
  • For color palette transfer, a simple reference image with your desired colors works well
  • You can combine two style references by using a text description alongside an uploaded image
  • Very realistic photographic styles tend to produce the most “professional” looking outputs for commercial use

Step 2: The Generation Interface in Detail

The Text Prompt Field (Enhancing Your Inputs)

Below the three image input panels, there is an optional text prompt field. This is where you add context that the images alone cannot communicate:

  • Specific poses: “product placed on its side”
  • Lighting details: “dramatic side lighting from the left”
  • Composition: “close-up macro shot”, “wide establishing shot”
  • Additional elements: “surrounded by fresh herbs and vegetables”
  • Mood: “cinematic atmosphere”, “cheerful and bright”
  • Negative guidance: “no text, no watermarks, clean background”

The most effective approach: Let the images communicate the visual essentials (subject, setting, style) and use the text field for specifics the images cannot show — composition, lighting direction, mood, additional elements.

Generating and Iterating

After setting up your inputs:

  1. Click Generate
  2. Whisk produces a grid of variations (typically 4 images)
  3. Click any image to see it larger and access refinement options
  4. Use Regenerate for fresh variations with the same inputs
  5. Use the Edit or Refine option to adjust specific aspects of a selected image
  6. Swap any single input (Subject, Scene, or Style) to explore variations while keeping others constant

The iteration strategy: Fix what works, change what does not. If you love the style and scene but the subject is being distorted too much, try a cleaner subject photo. If you love the subject and scene but the style feels wrong, swap only the style input. Whisk’s three-input structure makes this targeted iteration natural.

Step 3: Real-World Workflows

Workflow 1: Product Photography Mockups (E-Commerce and Marketing)

The problem: Professional product photography is expensive and time-consuming. You need your product shown in multiple lifestyle contexts, seasonal settings, and stylistic treatments.

The Whisk solution:

  1. Subject: Your product photo (clean image on white or simple background)
  2. Scene: A lifestyle environment photo (a kitchen counter, a coffee table, an outdoor terrace, a workspace)
  3. Style: A reference image from your brand guidelines or a commercial photography style you want to match

Generate and evaluate: Does the product look natural in the setting? Is the lighting consistent? Is the style match close enough?

Iterating for a product line: Once you have a scene + style combination that works, keep those inputs constant and swap only the Subject for each product in your line. This produces a coherent visual collection across multiple products.

Example prompt text to add: “Product centered in frame, clean composition, commercial product photography quality, no people”

Use case specifics:

  • Show a coffee mug in a cozy home office setting with a warm editorial photography style
  • Show skincare products in a minimal bathroom scene with a luxury beauty photography aesthetic
  • Show outdoor gear in an alpine environment with a documentary photography style
  • Show food products in a kitchen scene with a lifestyle cookbook photography look

Workflow 2: Consistent Character Across Multiple Scenes

The problem: You have a brand mascot, illustrated character, or product character that needs to appear in multiple different contexts while remaining recognizable.

The Whisk solution:

  1. Subject: Your character illustration or mascot image
  2. Scene: Swap this for each context you need (office, outdoor adventure, holiday setting, classroom)
  3. Style: Keep this constant (your brand’s illustration style)

The consistency advantage: By keeping the Style input constant, all generated images have a visual coherence even as the scene changes — which is difficult to achieve with text-only generation.

Practical output: A character that appears in four different seasonal scenes, all in a consistent visual style, without a separate illustration commission for each scene.

Workflow 3: Social Media Visual Series

The problem: Consistent visual identity across a series of social posts without a photographer or designer for each one.

The Whisk solution:

  1. Style: A reference image that defines your brand’s visual aesthetic (could be one of your best-performing past posts)
  2. Scene: The environment or setting theme for this series
  3. Subject: Swap for each post’s featured element (different products, concepts, or themes)

Example: A wellness brand creates a “seasonal wellness” series:

  • Style: consistent warm, soft photography aesthetic (one reference image)
  • Scene: varies by season (spring garden, summer beach, autumn forest, winter interior)
  • Subject: varies by post (yoga mat, herbal tea, journal, face serum)

All posts feel visually related without being identical. One style reference maintains the brand feel across a full series.

Workflow 4: Book Cover and Publication Concept Development

The problem: Authors, publishers, and content creators need cover concepts to evaluate before commissioning finished design.

The Whisk solution:

  1. Subject: The primary visual element of the cover (an object, a character sketch, a symbolic image)
  2. Scene: The atmospheric or environmental context
  3. Style: Reference covers from the same genre with the aesthetic you want

Generate multiple concepts rapidly: Explore 4–6 different subject + scene combinations while keeping the style constant. This produces a concept board in under 30 minutes that would take a designer days.

Important: These Whisk outputs are concept references for development, not finished covers. Take the concepts that resonate to a professional designer for execution.

Workflow 5: Interior Design Mood Boards

The problem: Communicating a design direction to clients or contractors requires visual clarity that text descriptions cannot provide.

The Whisk solution:

  1. Subject: A piece of furniture or decor item you want to show in context
  2. Scene: A room environment photo (similar to the client’s actual space if possible)
  3. Style: A reference image from the design style you are proposing (Scandinavian, mid-century, industrial)

Output: A generated image showing how a specific item might look in a specific space in a specific style — communicating the design direction more effectively than any separate image board.

Step 4: Advanced Techniques

The Style Transfer Experiment

A purely aesthetic exercise: take a single subject and apply 4–5 completely different style references to see the range of possible treatments.

Setup:

  • Subject: same image throughout
  • Scene: same image throughout
  • Style: swap through watercolor, oil painting, photograph, flat illustration, pencil sketch

Use case: When you are unsure which visual treatment is right for a project, this rapid exploration helps you identify the direction before committing.

The Scene Transformation Technique

Keep subject and style constant, systematically explore how the same subject looks across multiple scene contexts.

Practical example for marketers: A product shown in home, office, outdoor, and travel contexts — all in the same photographic style — to understand which lifestyle context resonates most with your brand positioning.

Text-Only Generation as a Starting Point

You do not need to upload images for all three inputs. You can type descriptions for any or all of them:

  • Subject: “a sleek black espresso machine”
  • Scene: “a modern scandinavian kitchen with wood countertops and white cabinets”
  • Style: “professional lifestyle commercial photography, warm morning light”

This makes Whisk function as a more structured alternative to standard text-to-image generation — the three-field structure guides you to think separately about subject, environment, and aesthetic.

Combining Text and Image Inputs

For maximum control, combine text descriptions with uploaded images. For example:

  • Subject: upload a product photo + text: “shown at a 45-degree angle”
  • Scene: upload an environment photo + text: “early morning, soft diffused lighting”
  • Style: upload a style reference + text: “slightly desaturated, cinematic color grading”

Free Tier Optimization Strategies

Strategy 1: Prepare Your Inputs Before Opening Whisk

Whisk’s daily generation limit means wasted generations on bad inputs are costly. Spend 5 minutes curating your input images before starting a generation session:

  • Crop subjects to center and simplify backgrounds
  • Choose scene images with clear environmental identity
  • Verify your style reference is distinctive and unambiguous

Strategy 2: One Variable at a Time

When iterating, change only one input at a time. If you change all three simultaneously and the output improves, you do not know which change made the difference. Systematic single-variable iteration extracts maximum learning from each generation.

Strategy 3: Generate in Batches, Not Individually

Whisk generates 4 variations per prompt. Use this: vary the text prompt slightly between batches rather than regenerating identical prompts hoping for a different result. “Product on left side of frame” vs “Product centered” vs “Product slightly out of focus in background” — three distinct text variations, three distinct batches, 12 total variations to evaluate.

Strategy 4: Save Inputs That Work

When you find a Scene + Style combination that produces high-quality outputs for your use case, save those reference images as templates. Label them in a folder (e.g., “Whisk — approved lifestyle scene”, “Whisk — approved editorial style”). Reuse them rather than finding new references each session.

Strategy 5: Use Whisk for Concepts, Imagen 3 for Finals

During Labs, Whisk is primarily a concept and exploration tool. For final-quality generation, the underlying Imagen 3 model is accessible through Google AI Studio at higher quality settings. Use Whisk to find the right direction, then use AI Studio for the final execution.

This section is not optional — it is genuinely important for anyone using Whisk professionally.

What You Can Use as Inputs

Clear to use:

  • Your own original photos
  • Images you have purchased appropriate licenses for
  • Royalty-free images from sources like Unsplash or Pexels (check their terms for AI training/generation use)
  • Screenshots of your own products, designs, or brand materials
  • Publicly available images you are using as style references (not as direct copies)

Requires caution or explicit permission:

  • Photos of identifiable living people (requires explicit consent — do not use photos of celebrities, colleagues, or anyone without permission)
  • Images from artists whose style you are referencing (currently a legally unsettled area — be aware of the ongoing intellectual property debates around AI style transfer)
  • Branded visual assets you do not own

Never use:

  • Images of real people without consent for any commercial purpose
  • Trademarked characters, logos, or brand imagery (Disney, Marvel, Nike, etc.)
  • Copyrighted images you do not have license to use
  • Images intended to deceive about a real person or event

Transparency About AI-Generated Images

For commercial use, consider whether your context requires disclosure that images are AI-generated. Regulatory requirements around AI-generated content disclosure are evolving — check current guidelines in your jurisdiction and industry.

For editorial, journalistic, or documentary contexts, AI-generated images are not appropriate as substitutes for real photographs of events or people.

Whisk vs. Other AI Image Generators

Feature Whisk Midjourney DALL-E 3 Adobe Firefly Stable Diffusion
Image-as-prompt input ✅ Core feature ✅ Limited ✅ Limited ✅ Style reference ✅ img2img
Three-way synthesis ✅ Unique ❌ ❌ ❌ ❌
Base model quality Imagen 3 (excellent) Excellent Very good Good Varies
Free tier ✅ During Labs Limited Limited Limited ✅ Open source
Commercial use ✅ Check current terms ✅ Paid plans ✅ Paid plans ✅ ✅
Google ecosystem ✅ Native ❌ ❌ ❌ ❌
No prompt engineering needed ✅ Images do the work Partial Partial Partial ❌

The verdict: Whisk’s three-input structure is genuinely unique. For users who think visually, have reference images available, and want controlled creative exploration, it is the most natural image generation workflow available. For users who are comfortable with detailed text prompts and need maximum precision, Midjourney or Stable Diffusion may still offer more control.

Common Mistakes to Avoid

Mistake 1: Uploading Cluttered Subject Images

A product photo with a complex background, a character against a busy scene, or any Subject image where the main element is not visually isolated will confuse Whisk. The model needs to clearly understand what the Subject is. Clean, simple, high-contrast Subject images produce dramatically better results.

Mistake 2: Using Mismatched Style and Scene Inputs

A hyper-realistic photographic style reference combined with a cartoon scene reference will produce incoherent outputs. Your style and scene should be compatible in their visual registers — both realistic, both illustrative, or explicitly contrasted as a creative choice.

Mistake 3: Expecting Exact Replication

Whisk generates new images inspired by your inputs — it does not composite or precisely reproduce them. If you need pixel-precise reproduction of your product or character, Whisk is not the right tool. For concept development and creative exploration, it is excellent. For exacting reproduction, use traditional design tools.

Mistake 4: Using Celebrity or Copyrighted Character Images

This should not need repeating, but the temptation is high. Using a celebrity’s photo as a Subject or a Disney character as a Subject is both a terms of service violation and potentially a legal issue. Do not do it.

Mistake 5: Not Iterating Enough

Single-generation sessions rarely produce the best results. Budget time for 3–5 iteration cycles per project. Each cycle teaches you how your specific inputs interact with the model, and the fifth generation of a well-developed prompt is often dramatically better than the first.

FAQ: Google Whisk

Q: Is Google Whisk free? A: Yes, Whisk is free during the Google Labs experimental phase at labs.google.com. Daily generation limits apply. Pricing may change if and when Whisk graduates from Labs to a full product.

Q: Can I use Whisk outputs commercially? A: Check Google’s current terms for Whisk at labs.google.com. Generally, AI-generated images from Google’s tools can be used commercially, but always verify the current terms as they are subject to change.

Q: Does Whisk store or use my uploaded images to train its models? A: Google’s standard Labs terms apply — review the specific terms for Whisk at labs.google.com. If you are uploading sensitive or proprietary images, review the data handling terms carefully before uploading.

Q: Why does my subject look distorted in the output? A: Whisk is synthesizing a new image, not directly placing your subject. Strong stylistic style inputs, complex scene backgrounds, or subjects without clear silhouettes can all cause distortion. Try a cleaner Subject photo and a simpler Style reference.

Q: Can I upload photos of my own face or my clients’ faces as Subject inputs? A: You can technically upload photos of people as Subject inputs, but Whisk’s terms specifically address this. For commercial purposes, only use photos of people with their explicit consent. For your own image, you can use personal photos of yourself.

Q: How does Whisk compare to Google ImageFX? A: ImageFX (also at labs.google) is text-to-image generation using Imagen 3. Whisk is image-as-prompt generation using the same model. They serve different creative needs: ImageFX when you can describe what you want in words; Whisk when you want to show what you want through reference images.

Q: What resolution are Whisk’s outputs? A: Output resolution varies and may be lower during the Labs phase than in a final product. For production use, check the current resolution specifications and whether outputs are suitable for your intended format.

Conclusion

Google Whisk solves a problem that every creative professional who has used text-based image generation has encountered: the exhausting gap between what you can see in your mind, what you can describe in words, and what the AI actually produces.

By making images the primary communication medium — showing the model a subject, a scene, and a style rather than describing them — Whisk dramatically reduces that gap for users who have visual references available. The three-input structure is not a gimmick; it reflects how creative professionals actually think about visual work.

The practical applications are immediate: product photography mockups without a photoshoot, character consistency across multiple scenes, social media visual series with coherent brand aesthetic, concept visualization for design and editorial work.

The tool is free during Labs. The daily limit means you need to be intentional with your generations — which, honestly, produces better work than unlimited generation would. Constraints force clearer thinking.

Your next step: Go to labs.google.com. Find one product, character, or object photo from your existing files to use as a Subject. Find one environment photo for your Scene. Find one aesthetic reference for your Style. Generate four variations and see what happens.

The first generation will probably surprise you. By the fifth, you will understand exactly what this tool can do for your specific creative work.

📚 Continue the Series:

Last updated: April 2026. Google Whisk is an experimental Labs product. Features, daily limits, pricing, and availability are subject to change. Verify current status at labs.google.com. Always review the current terms of service for commercial use before using generated images in commercial contexts.

⚠️ Do not upload images of real people without their consent. Do not use trademarked characters or logos as inputs. Review intellectual property considerations before using artist style references commercially. AI-generated images are not appropriate for journalistic or documentary contexts.

Related Articles

Disclaimer: The information contained on this blog is for academic and educational purposes only. Unauthorized use and/or duplication of this material without express and written permission from this site's author and/or owner is strictly prohibited. The materials (images, logos, content) contained in this web site are protected by applicable copyright and trademark law.