Tools
AI Product Photography Guide
Plan a product photo set around one approved reference. Choose the shots you need and reuse the same lighting and colour instructions across the set.
How to plan your first AI product photo set
Start with a clear photo of the item you sell. Choose one shot for its destination: a clean main image for a listing, or a lifestyle scene for a campaign. Write down the backdrop, camera angle, light direction, and details to preserve, such as the cap, colour, and label. Generate one draft, compare it with your reference, and keep the lighting and framing instructions for the rest of the set once the first shot passes review.
Packshot, lifestyle, or something in between
Three jobs get mixed up, and each has a different standard. Packshot: the product alone, centred, on a clean or transparent background, even lighting, no props. This is the listing image, and the standard is accuracy — the buyer is checking shape, colour, and what is in the box. Lifestyle: the product in a real setting with props, hands, and environmental light. This is the image that sells, and the standard is plausibility rather than precision.
And the hybrid that actually works best for most catalogues: the product clean and sharp, placed in a real environment with a soft shadow, no props touching it. Choose by where the image will be seen. A marketplace grid rewards consistency and accuracy; a brand page or a campaign rewards a world. A set that mixes the two without a rule looks like two different photographers were hired.
Keeping true product colour across a set
Product colour is the one thing a buyer will compare against the object in front of them, so drift is unacceptable even when nobody else would notice. The reason colour drifts is usually the lighting clause: the same product shot under "warm soft light" and "bright daylight" is two different colours, and across a set those differences accumulate. The fix is procedural. Write the colour description once — the actual hue, plus the finish, plus any secondary colour — and paste it identically into every prompt for that product.
Name the light as a source and a colour temperature rather than a mood: "neutral 5500K key light" rather than "clean bright light". And when a frame drifts, regenerate it against the approved hero rather than colour-correcting it, because correction shifts hue in a way that is hard to match on the next frame.
Materials: glass, brushed metal, matte plastic
Material is what makes a product shot look photographed rather than rendered, and each material fails in a different way. Glass: you need something behind it to refract, and you need the edge to catch a highlight — a glass bottle on a flat background with nothing behind it looks like a plastic outline. Say what is behind and through it, and ask for a bright edge on one side.
Brushed metal: anisotropic streaks are the tell, so describe the grain direction rather than saying metallic, and give it something to reflect, because flat metal with nothing to reflect reads as grey paint. Matte plastic: the failure is over-gloss, so say matte or satin explicitly and ask for a soft wide highlight instead of a sharp specular dot. In all three cases the same rule applies — material cues need an environment to reflect, which is why a pure white sweep flattens metal and glass more than anything else.
Shadows and contact points
The single detail that most often makes a generated product look pasted on is a missing contact shadow. A floating product reads as a sticker, no matter how good the surface rendering is. You need two things: a soft ambient shadow directly under the object where it meets the surface, and a directional cast shadow that anchors it to one light source. Ask for both by naming them — "soft contact shadow beneath the bottle, gentle cast shadow to the right".
Floating products and levitating arrangements are the hardest case, because there is no contact point; if you want a levitation shot, the shadow must stay on the surface below, softened, which is what actually sells the effect. And keep shadow direction identical across a set, because inconsistent shadows are the giveaway that several images came from different light sources.
Marketplace versus catalogue framing
Framing is decided by the placement. Marketplace listings want the product large in frame with a small margin, square 1:1, because the grid crops anything else and the buyer is comparing sizes across products. Catalogue and lookbook pages want air around the object, often 4:5 or 3:4, because the page supplies the spacing and the image is read as a composition. Advertising placements need 4:5 for feed and 9:16 for stories. Generate at the ratio you will place it at.
If one image must serve a marketplace grid and a catalogue page, produce it at 1:1 with generous margin so the 4:5 crop still holds the product with air, and check the crop before committing to a high-resolution pass. The other framing decision is camera height: slightly above centre for most objects, dead level for anything with a strong vertical.
What to check before you make more images
Put the first draft beside the original. Check that the outline and proportions match, that the colour still represents the actual product, and that every letter on the visible label is accurate. Preview the crop at the size buyers will see. Fix any differences in the first shot before reusing its prompt across the rest of the set.
Which engine for which product job
Nano Banana Pro is the default for product work because it holds a product's shape and its printed label steady across variants and renders soft studio light without being told twice — which is most of what a catalogue needs. Seedream is the one to use when the shot has to live in an environment: it handles surfaces, props and atmosphere better, and is worth it for lifestyle frames rather than for the listing image.
GPT Image 2 earns its place when the frame carries text — a price badge, a size label, packaging copy in shot — because it renders readable lettering where others produce decorative squiggles. The practical test is cheap: shoot one product on two engines with the same prompt and compare the edges of the label and the behaviour of the shadow, because those two details separate a usable catalogue image from a pretty one.
Lighting a product without a studio
A real studio has a large soft source, a fill, and a controlled background, and you can describe all three. The most transferable setup is a large soft key at roughly forty-five degrees and slightly above, a weaker fill from the opposite side to keep the shadow side readable, and a background that is lighter or darker than the product rather than the same tone.
Describe the light by source, size, direction and temperature, and skip mood words entirely: "large softbox from the left at forty-five degrees, soft fill from the right, 5600K" is a lighting instruction, while "bright airy commercial lighting" is not. For glass and metal, add a vertical strip light on the reflection side, which is what produces the long highlight down the edge of a bottle. Keep the lighting clause identical across the whole range and change only the product.
A SKU range in one sitting
Approve one shot to set the look for the range. Reuse its background, light source, camera height, and framing instructions. For each SKU, upload a reference of that exact item so the model can see its own shape, colour, and label. Review each result next to its corresponding reference before generating the remaining angles.
Where product shots go wrong
Four failures account for most rejected product images. Wrong colour: almost always the lighting clause drifting between frames rather than the product being misread. Unreadable label text: no engine guarantees glyphs, so if a SKU's printed copy is legally required to be correct, composite it from a real photo of the label rather than generating it, or shoot that one frame conventionally.
Melted geometry: a bottle or cap whose shape wobbles, usually caused by a prompt that stacks product features and props together — describe the product, then the prop, and never in one clause. And inconsistent shadows, which come from changing the light clause between frames. When one of these appears, the fix is nearly always to go back to the approved hero and regenerate the single frame, not to correct it, because corrections do not match the rest of the range.
Prompt structure for product stills
Product prompts work best in a fixed order, because the model weighs earlier words more heavily. Order: product and material, then finish and colour, then framing and camera height, then light by source and direction, then background, then what must not change. A workable example: "matte black aluminium water bottle, single product, centred, camera slightly above centre, large softbox from the left at forty-five degrees, soft fill from the right, seamless warm grey background, soft contact shadow beneath, cap on, no props touching the bottle, no text overlay."
Notice that the last two clauses are negative: for product work, what must stay clean and what must not appear is often more useful than another adjective. Change one slot at a time between attempts, or you lose the ability to tell which slot fixed the frame.
What a range costs in credits
Range cost is driven by retakes, and retakes are driven by whether the shared block was actually shared. Budget for one approved hero plus one batch: the hero costs a few drafts to find, and the batch should need roughly one re-roll per SKU rather than one rescue per SKU.
Where the range is large, the saving comes from resolution discipline — draft the whole range at standard quality, then pay for high resolution only on the SKUs that pass review, because resolution is billed per job and a range is mostly the same job repeated. If a lifestyle frame is planned, spend it on one hero SKU rather than on all ten, because the listing image is what converts and the lifestyle image is what explains.
And if the range will be re-shot next season, keep the shared block and the approved hero in your library: the second range is then a batch, not a project.
Open studio
FAQ
- Can it match my real packshot?
- Tight prompts plus optional references help. Always run brand and legal review before publishing.
- Is this a background remover?
- Ofrai generates full frames. Ask for a clean sweep in the prompt instead of a dedicated cutout tool.