AI + Industrial Design
Consistency in Gen AI design
A shuriken-shaped pasta experiment in moving from impressive one-off images to a repeatable visual workflow.
Generative AI is excellent at producing convincing individual images. The harder problem begins when the same product has to remain recognisable across several views.
To explore that problem, I designed a pasta for Render Weekly’s #RWPasta challenge and built a workflow across Ideogram, Vizcom, Meshy, Rhino, ChatGPT and KeyShot.
Explore
Use Ideogram to test forms, materials and lighting before choosing a direction.
Define
Turn the selected idea into controlled geometry and a reusable textured 3D asset.
Systemise
Separate shape, texture, environment and mood into a repeatable prompt structure.
Contextualise
Combine the same model, palette and scene logic across multiple final views.
A consistency experiment
I had mostly used generative AI for quick ideation. This time I wanted to push it further: could a set of tools carry one product from an early idea to multiple finished images without losing its identity?
The goal was not simply to generate a beautiful pasta image. It was to make the same pasta believable from every angle.
Exploring the initial concepts
I began in Ideogram, exploring pasta shapes, surface textures and lighting setups. The breadth was useful: it helped me identify the character I wanted before committing to geometry.
Defining the product in 3D
With a direction in mind, I needed a reusable 3D model. Vizcom struggled to interpret the complex topology. Meshy produced useful starting points from images and generated convincing food textures, but its geometry was not refined enough for the final design.
I therefore modelled the pasta manually in Rhino 7, then used Meshy to transfer the texture onto the controlled form.
The consistency problem
I moved the textured model into Vizcom. Individual images looked polished, but the object’s proportions, details and visual character shifted from one view to another.
I responded by treating the workflow like a small visual system rather than a sequence of isolated prompts.
Controlled form
The textured 3D model anchors proportion, topology and camera angle.
Structured prompt
ChatGPT separates instructions for shape, texture, environment and mood.
Shared palette
A custom Vizcom palette carries the same lighting and visual language between images.
Building a repeatable recipe
I returned to Ideogram and narrowed the project to two visual directions. I then used ChatGPT to make the prompts more explicit and created a custom palette in Vizcom from the selected references.
The combination—textured model, structured prompt and shared palette—finally produced a recognisable product across multiple angles.
Consistency improved when each tool was given one clear responsibility.
From product render to food context
Clean product views were only part of the story. I wanted a summery, dreamy setting that made the pasta feel edible and real. A quick KeyShot scene gave Vizcom the composition and scale it needed; the same prompt structure and palette supplied the atmosphere.
One object, many views
The final images hold onto the same central form while changing camera angle, scale and context. There is still room to improve the finer details, but the workflow is repeatable—and far more controllable than prompting each image from scratch.
Review
Consistency comes from constraining the right variables
Ideogram opened the search space. Rhino fixed the geometry. Meshy supplied texture. ChatGPT organised the instructions. KeyShot established context, and Vizcom brought the system together visually.
The process became reliable when I stopped expecting one tool to solve everything and instead built a chain in which every stage protected a different part of the design intent.
A consistent AI workflow is less about the perfect prompt and more about preserving the same decisions from one image to the next.
