If you sell on Amazon, Etsy, or your own store, you already know the routine: shoot the product, hate the background, retouch for an hour, repeat twenty times. AI has cut that loop down to minutes, but only if your prompts describe the right things. This guide walks through the workflow I use for AI product photos when a listing needs to look like it came out of a real studio instead of a render farm.
Start with the shot, not the product
The most common mistake is prompting a product like a catalog description. "Ceramic mug, white, matte glaze" gives the model almost nothing about lighting, distance, or surface. A photographer doesn't think in adjectives, they think in setups. So describe the setup first: where the camera is, how far away it sits, what the light is doing, and what the product rests on.
A workable prompt shell looks like this: "45-degree product shot of a matte white ceramic mug on a light oak table, soft window light from the left, subtle shadow under the rim, shallow depth of field, 85mm lens look, plain warm-gray background." That's four decisions — angle, light, surface, lens — and each one does real work.
Lighting is where fakes get caught
Buyers don't consciously analyze highlights, but they notice when something feels off. The tells are usually flat light with no direction, shadows that don't match the light source, and a product that floats because nothing grounds it. Fix all three by naming a light direction and a falloff. "Soft window light from the left" is enough. Add "a soft contact shadow where the product meets the table" and the object stops floating.
If the listing needs a pure white background for marketplace compliance, say so explicitly and keep the contact shadow. "Pure white seamless background, soft shadow beneath the product" reads as a studio shot. "White background, no shadow" reads as a cutout — which is sometimes right, but never for the hero image.
Give the model a material to hold on to
Materials are the second big failure point. Matte ceramic, brushed steel, woven linen, and frosted glass all behave differently with light, and generic prompts blur them together. Name the finish, not just the substance. "Brushed stainless steel with soft vertical highlights" produces something believable; "metal bottle" produces chrome soup.
For textiles, mention texture and drape. "Woven linen with visible thread texture, natural folds" keeps the fabric from turning into plastic. For glass, ask for "visible edge refraction and a soft highlight along the rim" and you'll avoid the milky look that ruins most AI glassware.
Match the set, not just the photo
One great image doesn't fix a listing. If you're generating a hero, two lifestyle shots, and a detail crop, they need the same light and color story. The simplest way is to lock the background and lighting language in every prompt and only change the camera position. Copy the same phrase block, swap "45-degree" for "overhead" or "macro detail of the lid," and you get a set that reads as one shoot.
Keep a plain text file with your three or four standard prompt blocks. When a new SKU arrives, you paste the block, change the product sentence, and generate. It takes about two minutes per image and, more importantly, the tenth product in a catalog looks like it belongs with the first.
Where it still goes wrong
Three problems show up over and over. Text on labels comes out garbled — generate the product clean and composite real packaging artwork later. Logos drift — keep them small or leave them off and add them in a photo editor. And scale gets weird — a coffee mug next to a "small plant" may come out the size of a barrel, so state relative scale when two objects share the frame.
None of these are dealbreakers. They're just the boundary between what a generator handles well and what still belongs in a normal editing app.
A quick workflow that holds up
Upload nothing, prompt everything, generate four variations, pick the one with the cleanest shadow, then check it at thumbnail size. Marketplace images are seen small first, so if the product doesn't read at 150 pixels, the lighting is probably too busy.
When you're ready to run a full listing set, you can generate it directly in the AI image generator and keep the prompt blocks handy between sessions. Studio-quality product photos stopped being a budget line item last year. The bottleneck now is just knowing what to ask for.