How AI Changes the Client Revision Round on a Display Concept

Ask a display manufacturer where the concept hours go and the answer is rarely the first render. The first render is the fun part. The hours go into round two, when the client wants the header taller, and round three, when they want to see it in a pharmacy instead of a supermarket, and round five, when marketing sends new artwork and procurement wants a front elevation for the quote. Each round is small. Each one reopens the whole piece.

I've worked in point-of-purchase for close to four years and I build AI POP Displays, so I've been on both ends of that thread. This post is about where AI actually changes the revision round, and it isn't the part most people expect.

Where do the hours on a display concept actually go?

Into the rounds after the first yes-or-no. A typical thread looks like this. The first concept goes out and the client likes the direction. Then the feedback arrives in pieces, from different people, in one email chain. The brand manager wants a bigger logo. The trade marketing lead asks how it looks in the actual store. Someone forwards the new campaign visual. The buyer asks for a version without product so they can see the structure. Procurement wants a straight front view.

None of those requests changes the concept. They are refinements of an image the client has already half approved. In a 3D workflow every one of them is still a trip back to the model: open the file, change the part, re-light, re-render, drop it into the store photo again, export, attach, send. The work isn't hard. It's just that five small requests become five sessions, and nobody bills for any of them because the order hasn't been placed yet.

That is the round AI should compress. Not the first image. The fourth.

Why does a revision in a generic image model start the design over?

Because a generic model has nothing to revise. There's no object under the picture, only pixels that were the most likely answer to the last prompt. When you rerun that prompt with "make the header taller" added, the model samples the whole scene again. The header grows, and the shelf spacing, the base width and the number of packs on the second tier move with it.

That is worse than a slow round, because the client sees it. They asked for one change and got four. Now round three is about why the shelves moved, and the person answering has to explain a change they didn't make. I went through the prompting side of this, feeding the approved image back and restating what stays frozen, in how to prompt AI for retail displays. It works when you do it carefully, every time, on every request, and anything you forget to restate is free to move.

The revision round needs the opposite default. Everything holds unless the request says otherwise.

What should a revision hold fixed?

The things the client has already approved without saying so. On a display that list is always the same:

• Structure. The format, the proportions, the number of shelves, the base.
• Materials. Corrugated stays corrugated, acrylic stays acrylic, with the same edges and thicknesses.
• Brand identity. The logo, the palette, the header artwork.
• Product load. Which packs, how many facings, where they sit.
• Light and camera. The same direction of light and the same viewpoint, unless the request is a new view.

That list is the default in our edit console. Every edit starts from an existing version, and the instruction that goes to the model carries those five things as fixed before it carries the change. The operation that asks for a new camera, or an empty display, or a line drawing overrides only the part it has to. The result lands as a new version in the same design, next to the one it came from, so the approved image is never overwritten and you can branch round four from round two if the client changes their mind.

If you have a concept in its third round right now, take it through the edit console on the free plan and send the client one change instead of a redraw.

Which client requests are a single edit?

Most of them, once you map the email to the operation. These are the requests I see most in a revision thread and what each becomes:

The client asks	The edit
"How does it look in the store?"	Change background, or convert to real life for an in-store photograph at shopper eye level
"Can we have it for the deck, in landscape?"	Change ratio, with the display kept fully in frame
"Show us the structure without product"	Remove product
"Put the new campaign visual on it"	Apply graphic, placed on the branded surfaces in perspective
"We need a front and a side view"	Generate angles, including true elevation, profile and plan with no perspective
"Send a clean drawing to engineering"	Convert to line drawing, same display, same angle, no color or product
Anything else ("slimmer header, warmer wood")	Custom edit, in your own words, with reference images attached if the change needs them

Each one is an edit on the image the client already saw, and at standard quality it costs a credit and comes back in about a minute. The views are the one that used to eat a day. A three-quarter concept that won the client's attention is not what procurement needs, and before this the flat views meant going back to the model to set up cameras on a piece that hadn't been ordered.

The composition matters more than any single edit. Round two used to be one session in the 3D file. Now it's a list of edits against the same approved image, run in the order the emails arrived, with every intermediate version still there.

When is it a new concept rather than a revision?

When the client changes the question instead of the answer. Three requests usually mean that:

1. A different format. The counter unit that becomes a floor stand. The proportions, the footprint and the product load are all different, so there is nothing to hold fixed.
2. A different material family. Corrugated to acrylic changes edges, thickness, joints and light. That's a new surface, not a tweak.
3. A different range. A new hero SKU, or twice the number of SKUs, changes what the fixture is built around.

For those, go back to the brief. In AI POP Displays the brief is saved with the project, so regenerating means changing the format or the material field and running it again, minutes rather than a redraw. The new concept starts a new line of versions, and the old one stays where it was in case the client walks it back.

The other boundary is millimeters. A concept render reads as correctly proportioned and nothing in it should be scaled off. Once the client approves the look, shelf pitch, board grade, joints and tolerances are engineering, done in the CAD your team already runs. I wrote about where that handoff sits in AI vs 3D rendering software for display concepts. The short version is that AI takes the speculative rounds out of the 3D seat, so the model gets built once, on a piece the client has already said yes to.

How should the client's answer come back?

On one link, not in a chain of attachments. The cost of a revision round isn't only producing the image. It's the thread around it: which file is the latest, who said what about which version, whether "the second one" means the second attachment or the second email.

In our workspace you pick the versions of a design you want the client to see and create a private share link. The client opens it without an account, looks at the versions side by side, and approves, requests changes or leaves a comment. The answer arrives in your workspace as a notification, attached to the design it's about. When the next round is ready you revoke the link and send a new one with the new versions, so there's never a stale link in someone's inbox pointing at round two.

The effect on the thread is that feedback stops arriving as prose you have to interpret. It arrives as a decision on the images the client just looked at. Approval is a click, and a request for changes comes back tied to the design it's about, which is the input for the next edit.

If the round you're stuck in is the one where six people reply to the same email, set up the share link on your next concept and let the answers land on the image.

What does this change about the week?

The first render was never the bottleneck. The bottleneck was that every small request cost a session, so manufacturers batched them, and batching turned five minutes of client feedback into a week of back and forth. When each request is an edit on the approved image and the answer comes back on a link, rounds two through five collapse into the same day the feedback arrives.

That changes what you can afford to offer. You can send three views instead of one, show the store scene before anyone asks, and answer the artwork change the afternoon it arrives. All of it without opening a modeling seat on a piece that hasn't been ordered.

AI POP Displays runs on Google's Nano Banana Pro. What holds a display steady across rounds is the layer on top: a format taxonomy with proportions, materials that behave, product load from the real packs, references labeled by role, and an edit console that treats everything the client approved as fixed. That's close to four years of watching revision threads, built into the tool once.

On the brand side, the revision round has a second half the concept doesn't reach, the step where an approved image turns into something a workshop can build. That's what we're building next. Bellto is an AI agent for brands and agencies that takes a POP campaign from brief to shop drawings. It proposes the campaign mix and materials with the budget split per store, designs every piece with you at real scale in parametric 3D, and 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. We're building it now, 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 the campaign you describe when you sign up sets your place. Pricing goes first to the people on the list, and there's no card.

For the concept and its revision rounds today, sign up free with 15 one-time credits and no card required, and run your next round of client changes as edits. 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.