Product Photos Without a Studio: AI Prompts for Sellers
I spent nine years art directing product shoots before I ever typed a prompt. Studio rental, a photographer day rate, a stylist, two rounds of retouching. For a client with four SKUs that was a five-figure line item and three weeks of calendar. So when people ask me whether AI product photography prompts can replace that, my answer is annoyingly specific: for about 70 percent of what a small seller actually needs, yes. For the other 30 percent, absolutely not, and I will show you exactly where the line sits.
Because here is the thing nobody selling you a prompt pack will admit. The model is not your problem. Your prompts are.
Most people open ChatGPT, type "professional product photo of my candle on a white background," get something that looks like a stock image from 2011, and conclude the tool is not there yet. The tool is there. You just briefed it like a client who says "make it pop" and walks away.
What this guide covers
- Why most AI product photos look fake
- The prompt formula I use for every single shot
- The six shots every listing needs
- Lifestyle shots, where the money actually is
- Keeping the same product across every image
- The compliance part nobody wants to read
- Platform specs cheat sheet
- Frequently asked questions
Why most AI product photos look fake
Pull up any bad AI product image and you will find the same three tells. The shadow does not match the light. The product floats a few millimeters above the surface instead of sitting on it. And the highlights are painted on rather than earned by an actual light source.
None of that is a rendering failure. It is a briefing failure.
When you write "studio lighting," the model has to guess. Studio lighting covers a beauty dish two feet from the subject and a six-foot octabox bounced off a ceiling, and those produce completely different images. When you say nothing about where the light comes from, you get an averaged, directionless mush. That mush is what your brain reads as fake.
Real photographers do not think in adjectives. They think in equipment and positions. One softbox, camera left, forty-five degrees, white bounce card opposite. Once you start writing prompts that way, the output stops looking generated and starts looking shot.
The second thing people get wrong: they describe the product and forget the surface. A product sitting on white seamless paper behaves differently from one on a white acrylic riser, which behaves differently again from a white cyc wall. Tell the model which one. The contact shadow is the single strongest realism cue you have, and it comes entirely from the surface you name.
The prompt formula I use for every single shot
I run every product prompt through the same five slots. Subject, surface, light, camera, exclusions. Miss one and the model fills the gap with its own average, which is exactly the look you are trying to avoid.
Subject. Be boringly literal. "30ml amber glass dropper bottle with a matte cream label and a black rubber bulb." Not "elegant serum bottle." The model does not know what elegant means to you.
Surface. Seamless white sweep, brushed concrete slab, wrinkled linen, travertine tile. Name it.
Light. Source, size, position, and what bounces it back. This is the slot people skip and it is the one that matters most.
Camera. Focal length, aperture, and height. A 100mm at f/8 from eye level reads as a catalog shot. A 50mm at f/2.0 from slightly above reads as lifestyle. Same product, completely different job.
Exclusions. No props, no text, no visible studio reflections, no hands. Whatever you do not want, say so, because the model loves adding a sprig of something.
Here is that formula as a working prompt. Swap the bracketed part for your product and it runs as is:
Run that, then run your old prompt, then look at them side by side. The difference is not subtle.
Left is "professional product photo of a serum bottle, studio lighting, high quality." Right is the prompt above. Same model, same session, about forty seconds apart. The only variable was how much of the shoot I actually described.
The six shots every listing needs
Three to eight images per product is the working standard across marketplaces, and higher priced items generally justify the higher end. But the count matters less than the coverage. Six shots, each answering a different question the buyer has not asked out loud yet.
1. White hero. The one that has to obey the rules. Clean, centered, no styling.
2. Three quarter angle. Rotate the product roughly 30 degrees and drop the camera slightly. This is the shot that gives the object dimension and stops it reading as a flat cutout.
3. Scale. Buyers are terrible at reading dimensions from a spec table. Put the product next to something universally sized, or in a hand, and the returns rate drops.
4. Detail. Macro on the texture, the stitching, the label print, the closure. This is where you communicate quality without a single adjective in the description.
5. In use. Someone actually using it. Not staged near it. Using it.
6. Back or alternate. Ingredients panel, reverse side, the second colorway. This shot exists to remove the last hesitation before checkout.
Which of those six do you currently have? Most sellers I look at have one and four, then wonder why traffic converts badly. Shot three is almost always the missing one, and it is the cheapest to fix.
Lifestyle shots, where the money actually is
White background shots get you compliant. Lifestyle shots get you bought.
This is also where AI earns its keep hardest, because a lifestyle shoot is the expensive one. Location, props, styling, a model if there is a person in frame. That is the shoot that used to eat the entire budget. Generating the environment costs you a well-written paragraph.
The trick is to brief the scene like a location scout rather than a decorator. Time of day, direction of the light, the angle it enters the room, what it hits on the way in. Morning sun at a low angle through a window on the left produces long soft shadows across a surface. That single sentence does more for realism than any amount of "warm cozy aesthetic."
Change three words in that prompt and it becomes a different season. Swap "low sun" for "overcast diffused daylight" and you get the quiet Scandinavian look. Swap linen for dark slate and the whole thing turns moody. One prompt skeleton, an entire content calendar.
If you would rather start from a structure that already works, the Mockups & Product Templates collection is built on exactly this pattern. Every prompt in there came out of real client work, which is why they name light positions instead of moods. The Luxury Skincare Product Mockups set is the one I reach for most for beauty and wellness SKUs, and Custom Floating Product Showcases handles the hero shot you put at the top of a landing page.
Keeping the same product across every image
Here is the honest limitation, and it is the one that will bite you.
Generate six images from six separate text prompts and you will get six subtly different products. The label drifts. The cap changes proportion. The amber goes slightly more orange in shot four. OpenAI itself flags that maintaining visual consistency for a recurring object across independent generations is not guaranteed, and every seller who has tried to build a listing set from scratch prompts has run into it.
Two things fix this.
Generate the set in one request. The current model can return up to eight images from a single prompt while holding the same objects and style across the set. So instead of six prompts, write one prompt that asks for the six angles. Consistency inside a single generation is dramatically better than across six.
Better: upload your actual product photo as a reference. This is the workflow I use for anything that ships to a real customer. Take one decent phone photo of the real thing in flat daylight, upload it, and ask the model to change only the environment and the lighting. The product stays your product. The background becomes whatever you need.
That last sentence is doing heavy lifting. Say it explicitly every time, because the model's instinct is to improve things, and "improved" is another word for "no longer the product you sell."
The compliance part nobody wants to read
Read it anyway. This is the section that keeps your shop open.
The rules across marketplaces in 2026 are less scary than the panic posts suggest, and they all reduce to one idea: what gets a listing suppressed is misrepresentation, not the fact that a machine made the pixels.
Amazon permits AI-generated and AI-enhanced product images provided they accurately represent the physical product and meet the standard image requirements. Background replacement, color correction and lighting adjustment are fine. Generating features the product does not have, shifting its actual color, or distorting its proportions is not, and that is true whether a model or a retoucher did it.
Etsy went further in January 2026. Sellers are required to tick the "I used AI-generative technology" checkbox, choose "Designed by" rather than "Made by" when the primary visual is machine-generated, and state the AI use in the description. Etsy also requires that listing images reflect the actual item being sold.
So the practical rule is simple. Use AI freely for backgrounds, scenes, lighting, mockups, concept work and marketing graphics. For the image a buyer uses to decide what is arriving in the box, work from a reference photo of the real thing and change only the environment. Disclose where the platform asks you to.
Is that more friction than generating six images from nothing? Yes. Is it less friction than a suspended shop? Also yes.
Platform specs cheat sheet
Get these wrong and the best prompt in the world will not save the listing.
Amazon. Main image on a pure white background at RGB 255, 255, 255. Product fills at least 85 percent of the frame. Minimum 500 pixels on the longest side, 2000 by 2000 recommended so zoom works, 10,000 pixels and 10MB as the ceiling. No text, logos or watermarks on the main image.
Etsy. 2000 by 2000 square is the current best practice. The first listing photo has to be at least 635 by 635, and anything smaller gets demoted in search. No white background requirement at all, and styled photography with props, color and real context tends to outperform the sterile version.
Shopify. 2048 by 2048 recommended. No background requirement, so your brand decides.
Notice the split. Amazon wants the product isolated and provable. Etsy wants it in a life. That is not a contradiction to work around, it is two different buying moments, and it is why the six-shot set exists. When you generate at 2K in a square ratio, you can crop for all three platforms from one set of files.
Frequently asked questions
Do I need a paid ChatGPT plan for this?
For commercial use, yes, that is the sensible route. Paid plans carry commercial rights on generated images and give you the 2K output and flexible aspect ratios you need for listing files. Free-tier output at lower resolution will not hold up when a buyer zooms.
Can the model put my actual label text on the bottle?
Yes, and this changed recently. Text rendering in the current ChatGPT image model is accurate enough for short label copy, product names and packaging type, which was not true a couple of years ago. Long paragraphs of ingredient text are still worth checking character by character before you publish.
Should I still hire a photographer?
For your hero SKU, your brand campaign, and anything where a person's face is the selling point, yes. What AI removes is the long tail: the fortieth product variant, the seasonal background swap, the four extra angles nobody had budget for. That is most of the work.
Why do my generated shadows look wrong?
Almost always because you named the light but not its position, or named the position but not the surface. A shadow is the product of a specific light in a specific place hitting a specific material. Give the model all three and the shadow fixes itself.
How many images should a listing actually have?
Three at absolute minimum, six to eight for anything above a low price point. More images correlate with better conversion on higher priced items, mostly because they close off the buyer's remaining doubts one at a time.
Start with one product and one prompt
Do not try to redo your whole catalog this week. Pick your worst-performing listing, the one with two blurry photos taken on a kitchen counter, and rebuild just that one. Six shots. Reference photo for the ones that show the product itself, generated scenes for the rest.
Then look at the listing next to your competitors. That is the whole test.
My last full product shoot as an art director took three weeks and cost more than a used car. The set I built for this article took under an hour, and four of those five images came out of a first pass. That gap is the actual story here, and it has almost nothing to do with the model being clever and everything to do with briefing it like a professional.
If you want a running start, every prompt in the Mockups & Product Templates collection is structured the way this article describes, pulled from real client jobs rather than written to look good in a listing. 3D Floating Product Box Mockups covers packaging heroes, Realistic Skincare Mockups With Hands covers the in-use shot most sellers are missing, and if you just want to see whether this fits how you work, there are free prompts to test first.
