Between the Hype and the Actual Desk
Almost every founder who contacts us now arrives with a few AI-generated product images. Sometimes they are striking. The recurring question is whether AI in industrial design makes the designer unnecessary, or whether it is one more tool in the box. The honest answer sits in the middle and depends entirely on which stage you are at: generative tools are excellent while you are still searching for a direction, and close to worthless at the point where a design has to become a manufacturable part.
That is not a criticism of the tools. It is a description of what they produce.
Where Generative Tools Genuinely Help
- Fast concept exploration. Instead of ten sketches a day, you can scan hundreds of form directions in an hour and recognize a visual language worth developing. This is a real improvement to the divergent phase of concept work described in design thinking in product development.
- Mood boards and shared vocabulary. Getting a founder and a designer to confirm they mean the same thing by "rugged but friendly" used to take a week of back-and-forth. An image board built in minutes ends a lot of arguments early.
- Variations on a chosen direction. Once a concept is selected, quick color, finish and proportion variants are useful for internal decisions and informal audience checks.
- Writing and summarizing. Turning a messy client conversation into a structured first draft of a design brief is genuinely good use of a language model, provided a human corrects it.
- Topology optimization in CAD. A different and much more mature family of algorithms: given loads and constraints, it removes material where stress does not flow. Unlike image generators, this output is real geometry and can be manufactured — after manual cleanup and verification.
Where They Fail
The central obstacle is simple: image generators produce pixels, not geometry. A beautiful picture of a product is not a 3D model. It has no dimensions, no wall thickness, no assembly, and no commitment to being physically possible. In practice we see the same failures repeatedly.
Unmanufacturable geometry
No draft angles, no sensible parting lines, sharp internal corners, wild thickness transitions, and features that could never be released from a mold. All of it collapses the moment it meets design for manufacturing review.
No engineering intent
The image does not know there is a PCB, a battery, a motor and four screws inside, or that the housing has to open for service, or that a USB port needs a flat land around it. Internal volume drives external shape in real products — not the other way around. Enclosure work starts from the component stack, as covered in electronics enclosure design.
Inconsistency between views
Two images of "the same" product will disagree about proportion, button count and vent placement. There is no underlying object, so there is nothing to be consistent with — which makes a set of generated views unusable as a specification for a modeler.
Blindness to ergonomics and compliance
Grip, body dimensions, actuation force, accessibility and safety requirements are not derived from attractiveness. They come from measured data and testing, as in anthropometry in product design, and a generated image will confidently show a handle no hand can hold comfortably.
Sales imagery versus faithful imagery
A generated picture shows a product that does not exist. A professional product rendering is built from the actual CAD model, so it depicts what will be manufactured — including the parting line and the screw boss. Confusing the two is how a crowdfunding campaign ends up shipping something visibly different from its images; our article on product rendering cost explains what faithful visualization involves.
Intellectual Property and Originality
Three practical cautions, offered as engineering practice rather than legal advice — Projects House is an engineering firm, not a law firm, and questions of ownership and rights belong with an attorney.
First, the ownership status of generative-model output is not settled the same way in every market, and US registration practice around works lacking human authorship is a live question. Do not build brand assets on an image whose status you cannot state confidently.
Second, models are trained on existing material and can produce a form uncomfortably close to a recognizable product on the market. Before developing a generated concept, run a deliberate visual comparison against current products and a design patent search in your category.
Third, if you intend to protect the appearance, originality and a documented creation process matter. Our comparison of utility patent vs design patent covers what a design filing protects and what it does not.
A Workflow That Uses AI Correctly
- Generate widely, commit to nothing. Use image tools to build a wall of directions and pick a language, not a product.
- Redraw by hand or in CAD. A designer sketches over the chosen direction, resolving proportion, parting lines and interfaces into something buildable.
- Model around the internals. Real geometry gets built around the component stack, with wall thickness, draft, fasteners and service access — the step described in sketch to 3D CAD model.
- Render from the model. Marketing images come from the CAD, so what buyers see is what ships.
- Prototype and test. Grip, size and usability get confirmed physically, never from an image.
What to Ask a Designer Who Uses AI
Using these tools is not a red flag; hiding behind them is. Reasonable questions: which parts of this were generated and which were drawn? Is there a CAD model behind these images? Where does the internal component stack live in this form? Have we checked the concept against existing products in the category? Clear answers indicate the tools are being used as an accelerator on top of engineering judgment rather than as a substitute for it.
From Image to Product
If you have a folder of generated images you like and no idea how to turn them into something that can be tooled, that is a normal starting point and a good one — the exploration is already done. Send us the images and a description through the contact form, and we will tell you which parts of that form survive engineering and which need to change.