Campbell Underwater Photography CRO Audit
A ten minute walkthrough of the virtual wall preview built for Campbell Underwater Photography, and what the product page still asks a buyer to imagine before they commit to a large acrylic print.
Watch the full 10-minute recording.
Pressing play loads the video from YouTube — you can allow it for every teardown under cookie preferences.
What we found.
The work is the strength here. Underwater photography printed large on acrylic or metal is bought for one particular wall, and this catalogue gives people something they will not find in a frame shop. The sizes and finishes already exist, and the question raised before this build began was exactly the right one: how does a buyer judge scale.
That question is still unanswered where it matters. On the product page a size is a pair of numbers, and nothing translates 24 by 36 or 60 by 40 into a wall the visitor recognises. The preview in this recording runs as a standalone example with placeholder products, so the shopper never meets it at the moment they are choosing what to buy.
The same gap repeats at every step after size. Finish is a word until a render finishes, so acrylic and metal compete on description alone. An eight foot wall and a seventeen foot wall get the same product photo. And when a realistic preview is requested it takes 30 to 50 seconds and returns a two megapixel file, so the proof lands into dead time.
The direction is to move this out of the demo and into the product page, and to treat the render as part of buying rather than a feature beside it: hosting the AI provider will accept, the scale step reduced to one tap wherever possible, and the wait given something useful to show. A print sells when the buyer can see it on their own wall.
Problem, opportunity, impact
The product page sells large acrylic and metal prints, but nothing on it shows what one looks like on a wall, so the choice between 24 by 36 and 60 by 40 is left to the buyer's imagination at the point of purchase.
With the preview on every product page, a visitor uploads their own room, sets the real wall width, and sees the exact photograph, size and finish hanging there before deciding, which turns the size question from a guess into a settled decision.
Fewer visitors stall on size, fewer emails arrive asking how big the print really is, and the choice between acrylic and metal gets made by looking at a wall rather than reading two descriptions.
Scored across five dimensions.
Five dimensions, each scored out of ten.
-
01
Homepage & PDP review Where attention leaks and buyers hesitate, section by section.
-
02
Mobile UX audit 70% of your traffic — judged on its own terms.
-
03
Trust & credibility check The signals that make first-time buyers convert.
-
04
Collection & discovery Filtering, merchandising and guidance from browse to buy.
-
05
Prioritized CRO roadmap Ranked by impact vs. effort, ready to ship.
Every issue, paired with its fix.
14 issues found, 16 fixes proposed.
Preview Lives Outside The Product Page
The wall preview was demonstrated on a standalone example page with placeholder images, not inside the product page where a visitor is choosing a photograph, a size and a finish.
Print Size Is Only Ever A Number
A shopper picking 24 by 36 or 60 by 40 has nothing on the page that relates those inches to their own wall, which is the concern that prompted this build in the first place.
Scale Depends On The Buyer Measuring First
The most accurate scale method asks for the real wall width in inches or centimetres, so the visitor has to leave the page and measure before the preview can be trusted.
The AI Estimate Carries A Margin Of Error
The third scale option lets the model judge the wall on its own, and the walkthrough notes there is room for error, so a preview can show a print hanging larger or smaller than it really will.
Render Takes Thirty To Fifty Seconds
Every realistic preview runs 30 to 50 seconds in the background, and during the walkthrough that time passed with nothing on screen except the wait itself.
The Generated Image Is Slow To Load
The file the AI returns is around two megapixels, so even after the render completes the before and after toggle sits waiting for the image to appear.
The Tool Only Runs On Local Hosting
The build works on a local machine and cannot go live on shared hosting, because the AI provider blocks shared hosting IP ranges, so the server question is still open.
Finish Is Text Until The Render Finishes
Acrylic, metal and frameless are picked from configuration fields, so the difference between a glossy acrylic and a metal print only becomes visible after a full render.
Object Marking Needs A Known Object
The second scale method depends on the shopper knowing the true length of something in the photo, such as a sofa at 60 inches, and many room photos contain nothing they can measure.
One Render Shows One Print At One Size
Each preview places a single photograph at a single size and finish, so a buyer weighing two images or two sizes has to run the whole upload and scale process again.
Demo Products Are Placeholder Images
The items in the walkthrough were stand in images rather than the real gallery, so the link between each photograph in the catalogue and the sizes and finishes it is sold in is not yet in place.
Nothing Keeps The Preview After The Visit
The finished render offers a download and nothing else, so a shopper who closes the tab loses the one image that had convinced them the print would fit.
Phone Capture Was Not Shown End To End
The walkthrough says a phone camera can supply the room photo, but every example used an uploaded desktop file, so the route most buyers would take has not been seen working.
Angled Walls Were Checked One At A Time
Perspective and long walls were tested by hand during the demo, including a 210 inch wall, so what the render handles well is known only from those individual passes.
Put The Wall Preview On Every Product Page
Move the preview onto the real product page, beside the size and finish selectors, so the visitor meets it while they are still deciding rather than after they have left.
Move The Render Onto A Dedicated Server
Host the generation on a server with its own IP instead of shared hosting, so the AI provider accepts the requests and the tool can run on the live site at all.
Lead With Marking A Known Object
Offer object marking as the first scale option, because a shopper can point at a sofa or a door without leaving the page to find a tape measure and come back.
Show The Wall Width The Render Assumed
Print the wall width used on the finished preview, so a buyer can see it assumed 96 inches and correct that in one step when the number is wrong.
Label The Visual Estimate As An Estimate
Mark the automatic estimate clearly as approximate and ask the shopper to confirm a wall width before the largest sizes, where being a few inches out is visible.
Fill The Thirty Second Wait
Replace the blank wait with a state that names what is happening and keeps the flat placement visible underneath, so the visitor has something to look at while the render runs.
Give A Fast Answer Before The Slow One
Keep the instant drag and place preview as the first answer on scale, so a shopper sees the size within a second and the AI render arrives as confirmation instead of the only output.
Serve The Render At Web Size
Deliver a compressed version of the two megapixel image for the before and after toggle and keep the full file behind the download, so the result appears the moment it is ready.
Let One Room Photo Try Several Prints
Hold the uploaded room and its scale, and let the shopper swap photographs, sizes and finishes against it without repeating the upload and measurement steps.
Render Acrylic And Metal Side By Side
Produce both finishes on the same wall in one pass, so the choice between acrylic and metal is settled by looking at two images rather than reading two descriptions.
Email The Finished Preview To The Shopper
Offer to send the before and after to an email address, which keeps the image after the visit and gives the gallery a reason to follow up with the right size already agreed.
Map The Real Gallery Into The Configurator
Connect each photograph to the sizes and finishes it is actually printed in, so the preview only ever offers a combination the gallery can deliver.
Recommend A Size Once The Wall Is Known
When the wall width is set, say which size suits it, so a visitor with a 120 inch wall is told a 60 inch print leaves roughly 30 inches clear on each side.
Test The Phone Camera Route On A Handset
Walk the phone path from camera to finished render on a real device, since the photo of a wall is most likely taken while standing in front of it.
Carry The Approved Preview Into Checkout
Attach the chosen size, finish and render to the line item, so the buyer sees the same image at checkout and the order confirms exactly what they approved.
Keep A Record Of What Buyers Preview
Store the wall widths, sizes and finishes people try, with their permission, so the gallery learns which walls and which prints its visitors are actually shopping for.
The version we'd build.
The teardown in 7 lines.
- Underwater photography printed large is bought for one particular wall, so scale is the whole decision.
- 24 by 36 is a number, and a buyer needs to see 24 by 36 on their own wall.
- A preview only sells prints when it sits on the product page, not on a demo page beside it.
- Three ways to set scale is generous, and the one that needs no tape measure should come first.
- Thirty to fifty seconds is a long time to stare at nothing, so the wait needs its own screen.
- Shared hosting will not run this render, and that is a hosting decision rather than a design one.
- Finish is a texture, not a word, so show acrylic and metal together instead of describing both.
Questions this audit answers.
A visitor uploads a photo of their own room, or takes one on a phone, then sets the real scale of the wall. They choose a photograph, a size such as 24 by 36, and a finish such as acrylic or metal. The tool then generates a realistic image of that print hanging on that wall, with a before and after view.
Three ways. The shopper can type the real wall width in inches or centimetres, they can mark an object in the photo and give its true length, or they can let the model estimate the wall on its own. The first is the most reliable and the third is the fastest.
It is a visual estimate, and the walkthrough is explicit that there is room for error. It works well enough for a first look, but a buyer deciding between a 40 inch and a 60 inch print should confirm the wall width before ordering. Showing the assumed width on the result makes that easy.
The realistic image is generated by an AI model, and that generation happens on a server after the shopper submits their room and configuration. The returned file is around two megapixels, so it also takes a moment to load. The fix is to show the fast flat placement immediately and treat the AI render as confirmation.
Yes, the flow supports taking the room photo directly with a phone camera. Every example in the walkthrough used a desktop upload, so the phone route still needs testing end to end on a real handset. It matters, because most people photograph a wall while standing in front of it.
It does. One of the examples was deliberately shot at a perspective angle on a long wall of around 210 inches, and the render placed the print with the correct spacing on each side. Those checks were done by hand, so the range of rooms it handles well is worth testing more widely.
The AI provider blocks shared hosting IP ranges because of past misuse, so generation requests from that kind of server are refused. The build works on a local machine today. Going live needs a server with its own IP, and the cost of that is a separate conversation.
Today they pick one finish, wait for a render, then change the finish and wait again. Rendering both finishes on the same wall in a single pass would let the shopper decide by looking. It also removes a second thirty second wait from the middle of the decision.
The shopper can download it, regenerate it, change the finish or try another size. Nothing saves it or sends it anywhere, so closing the tab loses the image. Offering to email the before and after keeps the proof with the buyer and gives the gallery a reason to follow up.
The walkthrough used placeholder products to demonstrate the mechanics, not the live catalogue. Each photograph needs connecting to the sizes and finishes it is genuinely printed in, so the preview never offers a combination the gallery cannot deliver.
The mechanics work, including scale, placement, finishes and angled walls. What is left is hosting that the AI provider will accept, the real catalogue wired in, and the preview placed on the product page itself instead of a separate example page. Those are the three things standing between this demo and a shopper using it.
The video runs about ten minutes. It walks through the wall preview flow as a shopper would meet it on a product page: uploading a room photo, the three ways of setting scale, choosing size and finish, placing the print, and the generated before and after. Three separate rooms were rendered, including a long wall shot in perspective, and the hosting constraint on going live is covered at the end.
See what's quietly costing you sales.
Tell us about your store and we'll send a personalized 5-minute video showing exactly where you're losing buyers — and how to win them back.
Takes about 40 seconds. No call required.
-
5-minute video teardownA personalized walkthrough of your store, recorded by Eva.
-
Homepage & PDP reviewWhere attention leaks and buyers hesitate.
-
Mobile UX audit70% of your traffic — judged on its own terms.
-
Trust & credibility checkThe signals that make first-time buyers convert.
-
Prioritized CRO roadmapRanked by impact vs. effort, ready to ship.