Generative AI product design is a goal-driven way of designing: you tell the software what a part must do (loads, mounting points, weight target, manufacturing method) and it proposes many candidate shapes that meet those requirements, instead of you drawing one shape and testing it. Engineers then compare the options, pick one, and validate it before building.
How generative AI product design works
In Autodesk’s Fusion implementation, the designer sets the geometry that must stay (for example mounting brackets and interface features), marks obstacle areas where no material may remain, adds manufacturing constraints and a target mass. The service then returns many valid options and reports how each performs against the requirements, per a DEVELOP3D review of Fusion generative design. Results can be compared side by side.
One caveat the same review raises: several vendors have used the term “generative design” to relabel older topology optimisation technology. Topology optimisation removes material from a defined block; Fusion’s approach does not need a starting volume. When you evaluate a tool, ask which of the two it actually does.
A real example: the GM seat bracket
The best-known case is General Motors and Autodesk’s seat bracket proof of concept. According to Autodesk’s customer story, the new part came out about 40% lighter and 20% stronger than the previous one. Gardner Business Media reports it replaced eight separate pieces with one 3D-printed stainless-steel part, and Modern Machine Shop reports the software produced more than 150 alternatives before engineers chose one on looks, manufacturability, simulation results and weight.
Two honest limits: this was a 2018 proof of concept, and sources differ on whether the 40% and 20% compare against GM’s own previous bracket or a conventional design.
How to use AI for product design: 5 steps
- Define the goal. Lighter, stiffer, cheaper, or a mix. Pick a measurable target such as a mass.
- Fix what cannot move. Bolt holes, interfaces and keep-out zones.
- Add manufacturing constraints. Process, material, minimum feature size. A shape you cannot make is not a result.
- Run the study and compare options on performance data, not looks alone.
- Validate and prototype. Simulate the shortlisted design, then test a physical part.
Generative design software compared
| Tool | Known for | Source |
|---|---|---|
| Autodesk Fusion | Cloud generative design that proposes many options with performance data; each study consumed cloud credits (a reviewer reported $25 per solve, so check current pricing) | DEVELOP3D |
| Altair Inspire | Physics-based generative design and topology optimisation, powered by Altair OptiStruct, with additive manufacturing tools | Altair |
| nTop (formerly nTopology) | Lattice structures and parts optimised for additive manufacturing; nTop 4 (May 2023) added programmatic workflow automation | VoxelMatters |
Feature lists and prices change often, so confirm them on each vendor’s site before buying.
Where it fits and where it does not
It pays off most on structural parts where weight matters and where additive manufacturing can make complex shapes, as in the GM bracket. It adds little to simple parts that are cheap to machine. The tool does not replace engineering judgement: someone still has to write the constraints, interpret the options and own the validation.
Related reading
- AI and human creativity in future industrial design
- The career of an AI prompt engineer for industrial design and simulation
FAQ
Is generative design the same as topology optimisation? They overlap. Topology optimisation works inside a defined block of material; generative design, as Autodesk describes it, can start without one and returns many alternatives. Vendors do not always use the terms consistently.
Do I need 3D printing to use it? No, but many results have organic shapes that are easiest to print, so add your real manufacturing method as a constraint.
Can it replace a designer? No. It explores options against the goals you set; people still set the goals and validate the result.