Can AI Design a POP Display? From Concept Render to Shop Drawings

Somebody asked me this on a call in July, and they asked it the good way. Not "is AI any good", but "can it design my display, yes or no, because I have a brief due Thursday". They ran design at a display manufacturer and had already tried a general image model on a real job. It gave them something beautiful and unbuildable.

The answer is yes, in two halves, and the split matters more than the yes. A display gets designed twice. Once as a concept, the visual a client approves, and once as a structure, the drawings a workshop cuts from. Almost everything people get wrong about this question comes from collapsing those into one, and a general image model gives you a picture of the first and nothing of the second.

I have worked in point-of-purchase for close to four years and I build AI POP Displays, which designs the first half. The second half is Bellto, which we're building now, and I get to it at the end.

What does AI POP Displays design on a display brief?

The concept, and the concept is five things.

It turns the brief into a visual. You paste the customer's email or write the job in a few lines, a briefing step fills the sector, format, material, style and retail environment, and a concept image comes back in under a minute. That used to mean a day of speculative 3D, or a careful hand visual, and a decision up front about which single direction to bet the hours on.

It holds format proportions, because the format is a structured input. A counter unit that comes back at counter height, an FSDU on a believable pallet footprint, a glorifier that holds one hero product at eye level without swallowing the shelf around it. In AI POP Displays the format comes from a taxonomy that carries its proportions. I went through the formats one by one in the guide to POP display types, and the short version is that each name carries geometry, not vibes.

It loads the real product. The pack shots you upload are placed as the product range, at scale and facing forward, with each variant kept separate, so the shelf count on the concept is a count of your packs and not of whatever balanced the picture.

It puts the fixture in the store. A supermarket aisle, a pharmacy counter, a boutique. That composited image is the one that gets a concept approved, because the client isn't buying an object on a white sweep, they're buying what their shopper sees.

It lets you show three readings of an ambiguous brief instead of guessing which one the client meant. This is the one that changes a week. Ambiguity in a customer email stops being a risk you absorb and becomes something you resolve by showing options. When the client picks one, every change after that is an edit that lands as a new version next to the original, so the direction they approved is never overwritten.

If your bottleneck is that you can only afford to put one direction in front of a client per brief, that is the specific thing our free plan unblocks.

What does the other half need?

This is the part the question is really asking. A concept image, from any model, settles the look and carries no geometry. Five things sit on the other side of that line, and each one is the job of a structure, not a picture.

Proof that it stands up. An image has no concept of load. It draws a shelf that reads as a shelf and can't know whether that cantilever carries a full facing of product, whether the base ships flat or whether the board survives a retailer's handling. It also has no failure state, so it returns a picture at the same confidence whether the geometry closes or not. Settling that takes a model with real dimensions, and that model is what Bellto designs every piece in, at real scale in parametric 3D.

Dimensions. A render isn't a measurement. There's no millimeter anywhere in it, so proportions can be believable and still not be the retailer's gondola end. In Bellto the proportions are derived from the product and the facings, so the numbers come from what the piece has to hold.

Production files. Cutting layouts and part lists have lived in structural CAD, operated by people who build displays for a living. I went through what each of those vendors shipped in AI display design, what actually exists in 2026, and the pattern was that nobody in structural CAD generated concepts and nobody generating concepts produced cutting layouts. Bellto is built to cover both ends. Every approved piece comes out as a 3D model, a part-by-part cutlist, dimensioned drawings, a STEP of the assembly and a cut DXF per part, and what a workshop does with each of those files is in AI shop drawings for POP displays.

A campaign, not one piece. A program is a floor stand, a glorifier, a shelf strip and a stopper that have to work as one campaign, spread across stores on one budget. A single image is one piece. Bellto starts from what you're launching (brand, product, channel, stores, budget, date), asks for what's missing and proposes the campaign mix and materials with the budget split per store.

The phone call. The one where the retailer changes the footprint two days before the deadline. On a picture that means a new picture, and on drawings done by hand it means a redraw. In Bellto a dimension change recomputes the whole piece and re-runs the manufacturing checks in seconds.

Why does this question return store interiors and TV panels?

Because the word display belongs to three bigger industries before it reaches ours, and searching for the answer shows you that immediately.

Most of what comes back renders whole rooms. Those tools turn a photo, sketch, floor plan or CAD file into a photorealistic store interior, and fixtures appear inside the scene as furniture. None of them takes one format, one material spec and one product line and hands back a unit a manufacturer could build, because that isn't the job they were built for.

The rest of the page is stranger. You get panel manufacturing engineering. You get UI design tools, because a display is also what appears on a monitor. You get design patent PDFs literally titled "Retail display". And you get a layer of retail execution vision, which points image recognition at shelves that already exist and reports what is wrong with them. Opposite end of the timeline, nothing in it draws anything.

What you almost never get is somebody answering the question for a physical fixture that has to be built at a real unit count. That gap isn't evidence of a hard problem. It's evidence that this category still doesn't have a settled name. AI POP Displays and Bellto are built for that fixture, and only for POP displays and POP materials.

What decides whether an AI concept is right?

Not the model. AI POP Displays runs on Google's Nano Banana Pro, and what decides the render is what sits on top of it.

Which questions settle a display before anything is generated. Which format name carries which proportions, so a counter unit isn't a floor stand with a shorter caption. Which materials behave a particular way, so a glorifier doesn't come back in corrugated. How many facings really fit given the packs you actually sell, loaded from the real packs rather than composed by the model. References labeled by role, so a pack shot is used as product and a sketch as structure, with sketch fidelity setting how closely the render follows the drawing. A briefing step that reads the customer's email and pulls those answers out before anything is drawn.

That's close to four years of this industry built into the tool, and it's the difference between the picture my caller got and a concept a client approves. The failure modes it removes are catalogued render by render in why generic image AI fails at POP displays.

Fifteen free credits is enough to run the brief that's actually on your desk through it and check the shelf count against your own packs.

How do you tell if an AI concept is worth anything?

One question sorts it, and it isn't a matter of taste. Could your manufacturer start the structure from this without rethinking the concept?

Three checks get you most of the way there. Does every shelf visibly land on something that carries it. Do the overall proportions match the format you actually asked for. Does the product load match how many facings really fit.

None of those is subjective, which makes them worth more than an opinion about which render looks nicest. A concept that fails one of them is a mood image, and mood images get a nod in a meeting and then get rebuilt from scratch by somebody at the factory who never saw the client's face. AI POP Displays is built around those three. The format brings its proportions, the load comes from your real packs, and on Pro a flat front elevation of the same display, with no perspective, is the view to count shelves on.

So, can AI design a POP display?

Yes, in two halves, and each half is a tool built for it. AI POP Displays designs the concept, today, from a brief, well enough that the speculative hours go away. The structure, the part with millimeters, parts and a cutlist, needs geometry, and a picture has none. That half is Bellto, and we're building it now. A general image model draws a picture of a display, and a picture is neither half.

Bellto is an AI agent for brands and agencies that takes a POP campaign from brief to shop drawings. You tell it what you're launching (brand, product, channel, stores, budget, date) and it asks for what's missing. It proposes the campaign mix and materials, floor stand, glorifier, shelf strip, stopper, with the budget split per store. Then it designs every piece with you at real scale in parametric 3D, with proportions derived from the product dimensions and facings, so a dimension change recomputes the whole piece and re-runs the manufacturing checks in seconds. Every approved piece comes out as a 3D model, a part-by-part cutlist, dimensioned drawings, a STEP of the assembly and a cut DXF per part, a package a workshop can quote without redrawing, and it suggests manufacturers that fit by material, format and volume. The quote still comes from the manufacturer. The waitlist is open to any brand or agency, and we're contacting the first ones soon to run the first real campaigns end to end. It opens by invitation, in small groups. Pricing goes first to the people on the list. No card. POP manufacturers can sign up too.

If the brief on your desk this week is a display, signup is free with 15 one-time credits and no card required. Pro is $49 a month for 150 concept generations at the founding price, with the $69 list price stated openly. Nothing you upload or render trains AI models, on any plan, because most of this work sits under NDA and that shouldn't be a paid feature.