How to Prompt AI for Retail Displays: Write a Brief, Not a Prompt

Most people meet AI for display work by typing something like "premium acrylic counter display for a skincare brand, photorealistic" into an image model, looking at what comes back, and deciding that AI can't do this. I understand the conclusion. The request did most of the damage, and the model never said so.

I've worked in point-of-purchase for close to four years and I build AI POP Displays. This is what I'd tell a designer at a manufacturer with a client brief in the inbox and a concept due.

Why does a short prompt fail on a display?

An eleven-word request has to be completed by something. The model has no failure state, so it never tells you the request was underspecified. It fills the gaps with whatever was statistically ordinary in its training data and hands you a picture at full confidence.

That's fine when the missing parts are decoration. It's expensive here, because on a display the missing parts are the deliverable. Format proportions, how corrugated board or acrylic actually behaves, how many facings really fit, whether the thing reads correctly in an aisle. Those are the parts a client approves and a factory quotes.

So the useful question isn't how to write a better sentence. It's what a display concept depends on, and the answer is almost always sitting in your customer's email already. That's a brief.

Which seven decisions does a display concept depend on?

The same seven a display designer answers anyway.

1. Sector. Cosmetics reads differently from beverage, which reads differently from pharmacy. Sector carries lighting, finish level, shelf density and the kind of store around the unit. It's one word, and it decides more than any adjective.

2. Display format. Counter unit, floor display or FSDU, endcap, shelf tray, glorifier, totem, window, shop-in-shop. These aren't synonyms, and each one carries real proportions. A general model reads the name as a style cue rather than a geometry, so the format you asked for comes back as a look instead of a shape.

3. Material. Not "premium" but E-flute corrugated, recycled board, acrylic, brushed aluminium, powder-coated steel, lacquered MDF, oak veneer. Materials behave differently under light and imply different thicknesses, joints and edges. To a general model a material is an adjective, a finish laid over whatever shape it drew. To a factory the material is half the shape.

4. Style. Minimal premium, industrial, sustainable, luxury, editorial, bold graphic. It's the decision the client reacts to first and the one with the least structure behind it.

5. Retail environment. White studio, soft grey studio, supermarket aisle, pharmacy, beauty boutique, storefront window. "Photorealistic" isn't a background, and if nobody decides, the model picks one. Studio when you're reviewing structure, store scene when the client is in the room.

6. Product load. How many SKUs, how many facings across, how many deep, how many tiers. Almost everybody leaves this one out, and it's the one that makes a render quotable. Without it you get the number of packs that balanced the composition.

7. Framing. Aspect ratio and camera position. A three quarter view at eye height sells a floor unit. A straight elevation is what you want when you're checking whether the structure closes.

The document you send a manufacturer covers more ground, and I wrote it up in how to write a POP display brief. These seven are the part a concept image depends on.

What did the vague request leave open?

Two cases from real briefs, with the names changed.

The first is the skincare counter unit from the top of this post. The request named a format, a material and a sector in one word each. The unit's proportions, the tiers, which bottles and how many, the header, the store and the camera were all open, so the model chose every one of them, and none of its choices came from the brand.

The second was a cardboard floor display for a soda brand in a supermarket, and that was the whole request. The customer's email behind it already had the shelf count, the facings per shelf, the can size, the board and the footprint. The request carried none of them, so all of it was the model's call.

In both cases the seven decisions existed. They were in the email, and they never reached the picture.

Does writing all seven down fix it?

It answers the questions on paper. Four things still go wrong when the request is complete, and most prompting advice skips all four.

The references get averaged. You can describe a bottle but you can't specify your client's bottle, so you attach the packshots, the logo, last year's unit and a mood image. To a general model that's a pile of pictures with no roles. It can't tell the mark to reproduce from the stand you only wanted the silhouette of, so it blends them, and the pack comes back as a cousin of the pack.

The revision round drifts. There's no object underneath the render. Each run is a new sample from a range your words only narrow, so when the client asks for a taller header, the shelf spacing, the base and the product load move with it. A complete request doesn't change that, because the drift isn't in the words. It's in the absence of anything to hold.

The brand mark comes back approximate. A logo isn't text to spell. It's a specific drawing in a specific color, sitting at an angle on a header or wrapped round a curved pack. Describe it and you get something near it. Describe it more firmly and you get a different something near it. That's a reproduction problem, and no wording reaches it. I went through the other ways this breaks in why generic image AI fails at POP displays.

You write it all again. The seven decisions are the same seven on every brief, and in a general model you supply them by hand, for every variant, every revision and every new customer. That's the cost nobody counts, a designer typing the same industry knowledge into a box, brief after brief.

So the seven decisions are the job. Writing them into a sentence is the part that doesn't scale and doesn't hold. If there's a customer email with a display in it on your desk right now, put it through the free plan and check the form it fills before anything renders.

How does AI POP Displays ask the seven for you?

It starts where the brief starts, in the customer's email. The briefing step reads the email and fills a form from a fixed display taxonomy, so whatever the email holds is in place before you touch anything, and you check a form instead of a paragraph.

Format comes from a list of display types that carry their proportions, so a floor unit starts out as a geometry, not a style. Materials come from a list that behaves. Styles, retail environments and aspect ratios are options, not words you hope the model reads the way you meant them.

Product load comes from the real packs. Upload the product photos and they're placed as the product range, not used as inspiration for something pack-shaped.

What changes is who asks the questions. The tool asks the same seven on every brief, and the knowledge of which seven matter is built in once instead of retyped by whoever has the deadline.

How are references and sketches handled?

Every reference travels labeled by its role, so the model knows which image is the pack to place, which is the mark to reproduce and which is only the look to borrow. We built it that way because unlabeled references were costing us renders.

Product photos are placed as the product range. The logo has two routes. At generation it travels as a labeled reference. On a render that already exists, apply graphic, which is a Pro edit, takes your artwork file and places it on the finished display. Either way, look at it at full size before the client does.

When the customer sends a drawing, you attach the sketch and set sketch fidelity. Faithful, balanced or free sets how closely the render follows the sketch's structure, silhouette and shelf count. The rest of that workflow is in sketch to render for POP displays.

How do revisions work without losing the design?

In a general model a revision is a new request. Here it's an edit on the render the client already saw.

The edit console works on an existing version. On the free plan you can change the ratio, change the background, or remove the product to judge the structure empty. On Pro you can generate angles of the same display, perspective views or a flat elevation, profile and plan, as a photo or a line drawing. You can convert the render to a line drawing or to real life, apply a graphic, or make a custom edit, an instruction in your own words with optional reference images attached.

Every edit lands as a new version next to the original, so the image the client liked is never overwritten. Compare puts two versions side by side, which is how you check that the shelves, the header and the packs are where they were before the change. The client-facing side of that round is its own subject, and I covered it in how AI changes the client revision round.

If round four is where your hours go, the free plan already includes three of those edits.

What can't a concept image carry?

Anything with a number behind it. A concept image carries proportions, not lengths, so millimeters, load, whether a fold locks and whether the unit fits a retailer's footprint aren't in it, however the request is written. Those live in a drawing.

AI POP Displays runs on Google's Nano Banana Pro. What decides the concept is the layer on top, and knowing which seven questions to build into it is what close to four years in point-of-purchase bought.

The drawing is the half we're building now. 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, and designs every piece with you at real scale in parametric 3D, proportions derived from the product and the facings. 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 comes from the manufacturer. The waitlist is open to any brand, and we're contacting the first ones soon to run the first real campaigns end to end. It opens by invitation, in small groups, and pricing goes first to the people on the list. No card. POP manufacturers can sign up too.

For the concept today, the seven questions get asked once and the image comes back in under a minute. Signing up 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. And if the job is the whole campaign, not just the picture, Bellto is where it goes next.