The Pitch Is "Stop Paying for Product Photography." Watch What It Does First.
There is a video going around aimed squarely at business owners, and the title is the whole argument: stop paying for product photography, use this AI instead. It is a good pitch. Photography is a real cost, the quotes come back higher than you expected, and every one of those images is out of date the moment you change the packaging.
Then you watch the demo, and the very first instruction is this: upload your product photo.
After that it adds a background, fixes the artifacts, sharpens the result and hands you something usable. Which is genuinely valuable work. It is just not what the title said. The tool selling you a replacement for photography opens by asking you for a photograph, and everything that makes the output good comes from the fact that a real object was sitting in front of a real lens at some point.
That gap between the pitch and the demo is the whole subject of this article, because it is where most small businesses are currently making an expensive decision badly.
Generating the Product Itself Is a Promise You Cannot Keep
We have written about the failure mode already, so this will be short.
When you generate the product rather than photograph it, you produce an image of something that does not exist. The weave is slightly wrong, the finish is glossier, the proportions have been quietly improved. The customer does not see your prompt. They see a promise, they buy, and then a box arrives containing the actual thing. The gap between those two moments is where returns and chargebacks live, and we covered where that quietly costs you in detail.
Take that as read. It is the premise of this article, not the argument. The argument is what you should be doing instead, because "don't generate your product" is not a strategy and leaves you with the same photography bill you started with.
The Ninety-Five Percent Nobody Puts in the Thumbnail
Here is the number that should reset your expectations before you spend anything.
Andre Kempy runs Admiral Media, a performance agency of about twenty people. He has done the thing the AI content tells you to do, properly and at scale: a systematized production line that pulls a brief and performance data from a document, generates hundreds of ad variations against it, and does it again the next day. Not dabbling. Infrastructure.
Asked on the Growth Masterminds podcast whether that generated creative outperforms what they were running before, his answer was: "I would say in 95% of the cases, no." The majority of his winners, he said, are still traditional footage of real people.
The second half of what he said is sharper, and it is about your business rather than your ads:
Read that again as an owner. Volume is not the same as progress. A pipeline that produces forty new ads a week produces forty new things to upload, forty new things to name, and forty new lines in a report, and every one of those activities feels like work being done. If none of them beat the ad you already had, you have bought motion. The question of who is actually making your ads, and what that costs, matters more than how many of them exist.
That is not an argument against using these tools. Kempy uses them daily. It is an argument for being precise about which job you are handing them, because the default job, make me the whole ad, is the one they are worst at.
Generate the Things That Do Not Have to Be True
So where is the line?
Fraser Cottrell's agency makes static ads for direct-to-consumer brands and he publishes teardowns of the ones that actually held spend. Look at what his winners have in common and the line draws itself.
One of his best performers is built around a cartoon pizza with a face on it, generated in about five minutes. His reasoning is purely practical: without that image the ad is scroll-past-able, and getting something like it drawn or 3D modeled the old way would have taken days. Nobody is going to receive that pizza in the post. It does not have to be true.
On another winner, a protein brand, the product is a real photograph, chosen because it is genuinely vibrant on camera. The review text is real, lifted from their own site. Only the small portrait beside the review is generated. On a soap brand, the bars themselves are photographed, because as he puts it, the actual bars look fantastic on camera and customers already recognize them.
That is the working rule, and it is simple enough to apply in a second:
The test is one question. Could a customer ever hold this image up against the real thing? If yes, it needs to be a photograph. The product, your premises, your team, your van, the finished job. If no, generate away without a second thought. Mascots, illustrated backgrounds, abstract scenes, stylized props, the impossible shot of your product floating in space.
Most of the AI creative you see failing is failing because somebody put the product on the wrong side of that line. The most expensive version of that mistake is putting a person on the wrong side of it, because a viewer who clocks a fake face stops believing the rest of the ad too.
What Editing Actually Means, in Practice
This is the part that gets skipped, because it is less exciting than generation and considerably more useful.
You already own photographs. Somewhere there is a folder from a shoot you paid for, plus a few hundred images on your phone. In almost every small business that folder is drastically underused, because the assumption is that a photo is a finished thing rather than raw material. Here is what these tools are genuinely good at doing to it.
Reframing without losing the product. A photo shot horizontally is useless as a story and awkward as a square. The old fix was cropping, which throws away the sides of your product to fit. The new one is generative extension: the model paints plausible background outward so the frame changes shape and the product stays whole. One shoot becomes every placement.
Relighting a flat photo. The single most common defect in owner-shot photography is not composition, it is light. A phone photo under a shop's overhead lighting looks gray and dead. Lifting it into something that looks intentionally lit is now a routine edit rather than a reshoot.
Changing the background, keeping the product. The same photographed item on a summer table, a winter counter, a workshop bench. Six seasonal ads out of one photograph, with the product identical in all of them because it is the same pixels every time.
Removing what you did not notice. The power cable, the reflection of you holding the phone, the cluttered shelf behind, the wrong-season decoration. This is the least glamorous item on the list and probably the one that will rescue the most images you already have.
Rescuing old assets. Catalog photography from years ago that was fine for print at the time and looks soft on a phone can usually be upscaled back into usefulness. That is a free archive most businesses have written off.
Every one of those operations shares a property that generation does not have: the product in the final image is the product that was photographed. You are changing the context, not the claim.
Your Shoot Is Worth More Than You Paid For It
This changes the economics of photography rather than eliminating them, and the change runs in your favor.
If one session yields eight usable images, a day's shoot is expensive per image and you ration it. If one session yields eight images that can each be reframed into four placements, relit, and dropped into three seasonal contexts, the same day produces something closer to a library. The cost did not move. The yield did.
Which means the honest advice to a small business right now is close to the opposite of the trending pitch. Do not stop paying for photography. Shoot less often, shoot better, and get far more out of each session. Photograph the things that have to be true, once, properly, with an eye on what you will need to expand them into later. Leave room around the subject. Get the product on clean backgrounds as well as styled ones. Shoot the same item from three angles.
If you would rather hand that whole loop to someone, that is the work we do: a human directing what gets shot and what gets generated, rather than a tool deciding for you.
Where the Seams Show
The failure mode of editing is subtler than the failure mode of generation, so it is worth knowing what to look for before you approve anything.
The most reliable tells are at the boundaries. Look at the edges of the product where it meets the new background, which is where extension work goes soft or leaves a faint halo. Look at shadows: if the light was coming from the left in the original and the new background is lit from the right, the composite reads as wrong even to people who could not tell you why. Look at hands, still the most likely thing in any frame to come out subtly deformed. Look at reflective surfaces, because a model will happily paint a reflection that does not correspond to anything in the scene.
Then look at text, and treat it differently from all of the above.
If an edit has altered your packaging text, the ingredient list, a certification mark, a size or a price, that is not a rendering artifact you can accept in a hurry. It is an inaccurate claim about your product, published at scale.
This is the point where the practitioner videos get quietly instructive. Watch the people who advocate pure generation review their own output and you will hear them reject it constantly. It looks staged. The lighting looks fake. That patient looks fake. And their fix, nearly every time, is the same move: go back and upload real reference images of the actual place. Even the workflow built to avoid photography keeps reaching for it to stop looking generated.
Keep the Originals
One piece of housekeeping that costs nothing now and is painful to fix later.
Keep the untouched files. Whatever you shoot, keep the originals separately from every edited version, and make sure you actually hold them rather than a freelancer, an agency or a tool's cloud account. Edited images are derivatives, and derivatives age badly: the model you used will change, the platform's preferred dimensions will change, and in two years you will want to redo an edit from the source rather than re-edit an edit.
The same goes for anyone you commission. Make it explicit that you are buying the raw files, not only the finished exports. It is an unremarkable request that gets forgotten constantly, and the businesses that forgot it are the ones who cannot re-cut their own library when something changes.
Start With What Is Already on Your Phone
Nothing here needs a budget approval. Do this version of it this week.
Open the folder from the last shoot you paid for, and the camera roll on your phone, and pick the three best photographs of the thing you actually sell. Not the most stylish, the clearest.
Take one of them and produce every placement you need from it by extending the frame rather than cropping it. Square, vertical, wide.
Take the second and clean it up. Fix the light, remove whatever is in the background that you stopped noticing months ago.
Then take the third and change its context to whatever season or setting you will be selling into next, and put it in the calendar for when that arrives.
You will have spent an afternoon and produced more usable creative than a generation pipeline would have given you in a week, and every image will show a customer the thing they are actually going to receive.
We do this for small businesses all day, and if you would rather have someone run it while you get on with the work, that is what we are for. If you would rather run it yourself, the list above is the whole method. Either way, do not throw away the photographs. They are the part nobody else can generate.
