You've generated a usable picture of a lamp, a chair, or a rug. Now you have to put it on a product page. That's where a lot of sellers lose the plot: they paste the image, skip any cleanup, and wonder why the listing looks flat next to a shop that clearly knows what it's doing. The image itself is rarely the problem. The gap is almost always in preparation and placement, not in how clever the prompt was.
If you're still deciding which tool to use for the base image, start with the AI image generator on the homepage and generate a few candidates there before you spend time on finishing work. Generating three options takes less time than rescuing one bad one.
Start from the product's real proportions
An image model will happily draw a table that is two metres long and forty centimetres high because nothing in the prompt told it otherwise. Before you write anything, measure your actual product. Width, height, depth. Then put those numbers into the prompt in plain language: "a low oak side table, sixty centimetres wide, forty centimetres tall, viewed straight on at eye level".
This one habit removes a whole category of rework. A buyer who makes furniture or sells parts notices wrong proportions instantly, and once they do, they stop trusting the rest of the page. If you can't give real numbers, at least reference something familiar — "about knee height", "roughly the width of a laptop" — so the model has a scale anchor.
Write the light before the object
Most people describe the thing and hope the lighting sorts itself out. Do it the other way round. Decide where the light comes from, how soft it is, and what colour it is, then describe the object inside that setup.
For marketplace listings, a softbox from the upper left plus a subtle fill on the right reads as "professional product photo" to almost everyone. Hard midday sun reads as a snapshot. Neither is wrong, but one sells furniture and the other sells honesty about the weather. Pick deliberately, and mention the shadow direction in the prompt if you want the object to look like it belongs in the frame.
A short prompt skeleton that holds up
Subject and material, then proportions, then lighting, then camera position, then what's excluded. Something like: brushed steel kettle with a matte black handle, medium size, soft diffused light from the left, three-quarter front view, plain warm grey background, no text, no logos. Five pieces of information. Not twenty.
Adding more adjectives past that point tends to make the image mushier, not sharper. If you need something fixed, name it once and clearly. Repeating "highly detailed, ultra sharp, 8k" five times does nothing except make the prompt harder for you to edit later.
Generate more than you need, then choose
One image is a gamble. Four is a decision. Even at the same prompt, small variations in how the tool interprets a word will give you different silhouettes, different shading, and occasionally a genuinely better composition than what you had in mind. Generate a small batch, put them side by side on a plain background, and pick the one that reads fastest at thumbnail size. The one that looks best at full size is often not the one that wins in a grid.
Keep the ones you reject. A pose you didn't want for this listing may be exactly right for a different product next month, and regenerating from scratch costs more time than renaming a file.
Fix the background before you resize anything
Tidy the edges, remove stray shadows, and straighten the horizon while the image is still at full size. Every crop and resize you do loses a little information, so all the fussy work should come first. A half-hour spent on a clean subject edge at 2000px is worth more than an hour of patching at 800px.
Also decide what the background is for. A product page can take a plain studio grey. A social post often wants a hint of context — a warm wall, a wooden surface, a corner of a room. Generating the same object twice, once neutral and once contextual, saves you a second round of prompt writing later.
Match the whole set, not each image
Sellers usually notice this too late. The first eight images look fine individually, and then a buyer scrolls the grid and sees eight different products from eight different brands. The fix is boring: keep a written record of the settings that worked — light direction, background colour, camera height, focal feel — and reuse it for every item in the range.
If you're generating a dozen items, write the shared half of the prompt once and only swap the subject line. Consistency comes from repeating the boring parts, not from luck. When one item refuses to cooperate, fix it in your editing app rather than regenerating from scratch with new wording, which will drag the rest of the set off-brand.
Check the file before it goes live
Export at the size the page actually needs, keep the background colour as close to your site's as you can, and look at the result on a phone. Most buyers will first see your product in a narrow column on a small screen. If the object has to be squinted at there, no amount of detail in the full-size file will save the listing.
You can run that final check in under a minute: open the generated image on the same device you'd shop from, at the same size it will display. If the shape and material still read clearly, the image is done.
None of this is complicated, and none of it is about writing a cleverer sentence. Measure the product, decide the light, clean the edges, repeat your settings across the range, and check it on a phone. If you want to test a few of these ideas on your own catalogue, the best AI image generator we've tested is a reasonable place to start a first batch.