Sketching with Words: Exploring AI in the Early Stages of Industrial Design

AI-generated imagery has been circulating in design studios for some time, often serving as a source of ambient inspiration or fuel for moodboarding. However, with the advent of more precise, prompt-driven tools like ChatGPT's image creator, we’re starting to see a shift: not just in what these tools produce, but in how they can be utilized to think, explore, and evolve early-stage industrial design ideas.

A New Medium for Conceptual Thinking

The big difference between tools like ChatGPT and more familiar visual generators like Midjourney is structure. Where Midjourney excels at aesthetic randomness, ChatGPT introduces a more deliberate approach. You can sketch with words. You can build an object iteratively. You can go back and revise specific details, like a conversation.

Rather than using AI to generate one-off images, we’ve started using it more like a collaborative sketchpad. You describe what you want to see—starting with basic shapes or materials, or interactions—and the tool gives you a visual. Then you edit it: change the color, add a bumper, swap the background. That back-and-forth can last minutes or hours. However, the image becomes more refined over time, aligning more closely with what’s in your mind.

A Step-by-Step Guide to Working with ChatGPT as a Visual Tool

To help other designers get started, here’s a breakdown of three core workflows we’ve tested, along with example prompts and resulting images. These examples use a new product concept for a compact countertop air purifier—a familiar object that benefits from exploring both form and surface detail.

(The Slide below generated with Descriptive Prompting)

1. Incremental Prompting (Build from Basics)

This is our most effective method. Start with primitive shapes and gradually build up complexity.

Prompt 1: "Create an image of a white cylinder on a black tabletop in soft daylight."

Prompt 2: "Add a perforated matte grey grille to the top surface and a subtle LED ring around its base."

Prompt 3: "Add a small digital screen on the front showing air quality levels. Include soft touch control buttons below the screen."

Prompt 4: "Place the device in a realistic kitchen environment with other neutral appliances."

This approach allows for high levels of iteration while keeping visual memory intact across steps.

2. Descriptive Prompting (Build from Imagination)

Use this when you have a strong mental image but no source file.

Prompt: "A compact countertop air purifier shaped like a vertical pebble, made of sandblasted aluminum with an ambient light ring. It sits near a potted plant on a white kitchen counter in morning light."

This works best for quick one-offs or when exploring overall vibe and design language.

3. Image Referencing and Editing

This is useful when you already have a visual starting point—like a sketch or render—and want to make quick visual changes.

Upload CAD render of cylindrical device.

Prompt: "Change the surface from smooth white plastic to speckled ceramic. Add fingerprint sensor to the back rim."

We took the 1st Image from the Incremental Prompting Method and Used it to mix with the prompt above. Additionally you can use source images with CMF you would like to apply.

Follow-up Prompt: "Re-render in moody evening light with a glowing blue LED ring and no background."

This flow is especially helpful in refining details or preparing visuals for reviews.

Where It Fits (and Where It Doesn’t)

We’ve found this tool most useful in the early phases to generate sacrificial concepts—the earliest part of the process where we explore a broad range of ideas without worrying if some are half-baked or off track. At Whipsaw, this typically happens right at the beginning of a project, when we’re framing the problem, identifying opportunity spaces, and generating visual material for conversation. We might use AI-generated images in a kickoff deck to help define the boundaries of a concept space or to elicit feedback from a partner before any commitments are made.

These images are visual hypotheses, and we now have a tool that can make a napkin sketch look like a photograph. ChatGPT allows us to revise specific visual attributes—such as color, material, proportion, or UI placement—with conversational precision.

But it’s crucial to be intentional. AI can generate a visual quickly, but it won’t evaluate whether that image makes sense for the user, the brand, or the use case. That’s still the designer’s job. In our process, designers guide and curate what is shown, what is discarded, and what is reworked using traditional tools. Sometimes, we reference these AI-generated visuals directly in mood boards or strategy decks. At other times, we translate them into CAD, sketch, or KeyShot to carry the idea forward in a more refined way.

That said, the tool is far from perfect.

A Tool, Not a Threat

More than anything, this has shifted how we think. Instead of jumping into CAD or Illustrator, we can now visualize early ideas faster. We can explore new directions for existing concepts without over-polishing. We can test speculative ideas that might not be worth modeling just yet.

But this doesn’t diminish the role of the designer. If anything, it reinforces it. The tool can generate, but it doesn’t decide. It doesn’t judge.

In that way, it’s not just another production tool. It’s a new kind of sketching medium. One that lets us build, question, and evolve ideas through language. That may be uncomfortable, even risky. However, it’s also an opportunity to rethink what the earliest phases of design can look like—and who or what we collaborate with to achieve this.

We’re still experimenting. But for now, the promise isn’t automation. It’s augmentation. And with the right mindset, it’s worth exploring.

‍