Online Product Photography with AI
The studio, the lighting, and the backdrop are the cheap parts to fake. The product itself is not — and almost every guide on this topic gets that backwards.
August 1, 2026 · 7 min read
Contents
Quick answer
Online product photography with AI means uploading one honest photo of the thing you sell and having a model rebuild everything around it — backdrop, surface, light direction, shadow, props. You still have to take that one photo. On Shotari's product photo editor the free built-in model does this at zero credits, and the paid models start at 3. What no tool does is generate a product you have not photographed: it will produce something plausible and wrong, and you will ship a listing image that does not match what arrives in the box.
One packshot is the whole input
Phone on a windowsill, product on a sheet of white paper, no flash, shot straight on. That is enough. Sharp focus and honest colour matter far more than the backdrop, because the backdrop is the part being thrown away.
What AI cannot do for you
Every one of these is a real failure mode, not a caveat added for balance. They are the reasons an AI product shoot goes wrong, in the order they usually bite.
- ⚠
It cannot invent your product
Prompting "a matte black water bottle with a bamboo lid" gets you a water bottle. Not yours. The proportions, the seam lines, the exact shade of black — all invented. Anything a customer will compare against the parcel they open has to come from an uploaded photo.
- ⚠
Label text and logos come back mangled
Small type on packaging is where image models still break. Ingredient lists, dosage panels, and wordmarks regularly return misspelled or subtly redrawn. Zoom to 100% on every generated label before it goes anywhere near a listing — and if the label is the selling point, keep the original crop and let AI change only the surroundings.
- ⚠
Free output is watermarked, so it is not listing-legal
Shotari's free tier — the 20 credits at signup, the 10 a day, and the 2 daily generations you get without an account — stamps a watermark on the result. Marketplaces including Amazon reject main images carrying watermarks or added graphics. Free output is fine for testing compositions; the image you actually upload has to come from paid credits, which are watermark-free and carry commercial rights.
- ⚠
Regulated categories are not a prompting problem
Supplements, cosmetics, medical devices, and food make claims through their imagery. A generated scene implying clinical testing or a health outcome is a compliance problem no amount of prompt tuning fixes. Keep those shots literal.
What a product shot costs
Costs are per generation at standard 1K resolution. Free credits — the signup 20 and the daily 10 — only run Shotari's own two models; every third-party model deducts from paid credits only, so a balance made entirely of free credits reads as insufficient the moment you pick one.
| Model | Credits at 1K | Where it earns its price |
|---|---|---|
| Shotari Basic | 0 | Free. Composition tests and throwaway variations. |
| Shotari Pro | 2 | Free-credit eligible. The cheapest usable listing image. |
| Seedream 5 Lite | 3 | Cheapest paid model. Good clean-background work. |
| Nano Banana | 4 | Fast scene generation when the product is simple. |
| GPT Image 2 / GPT Image 1.5 | 5 | Best odds on legible packaging text. |
| Nano Banana 2 / Seedream 4.5 / FLUX.2 | 6 | Realism for hero and lifestyle shots. |
| Nano Banana Pro | 10 | Highest consistency across a full product set. |
Two multipliers sit on top, and one of them quietly works in your favour. Resolution scales the cost — 2K is 1.5× and 4K is 2×, so a 4K hero from GPT Image 2 is 10 credits rather than 5. Aspect ratio also scales it, but only for 16:9 and 9:16, which cost 1.1×. Square, 4:3, and 3:4 are all 1.0×.
That is worth stating plainly because it lines up with the job: marketplace product images are square. Shooting 1:1 means you never pay the aspect surcharge at all — the 10% only applies to the widescreen and vertical formats you would use for ads and social, not for the listing itself.
Prompts that hold the product still
The pattern that works is the same one that runs through uncropping: describe what changes, never re-describe what stays. The moment a prompt narrates the product, the model treats the product as something to render rather than something to preserve, and it starts drifting. You can attach up to three reference images, which is the practical way to lock an angle or a colour you cannot put into words.
Marketplace main image
Replace the background with a pure white studio backdrop. Keep the product exactly as photographed. Add a soft contact shadow directly beneath it. No props, no text, no reflections.
Premium hero
Place the product on a polished marble surface with warm directional studio lighting from the upper left. Shallow depth of field, dark neutral background falling off behind. Product unchanged.
Lifestyle context
Place the product on an outdoor café table in soft late-afternoon daylight, one blurred plant in the background. Aspirational but uncluttered. Do not alter the product's shape, colour, or labelling.
Flat lay set
Top-down flat lay of the product with two complementary props arranged loosely around it, white background, soft even shadows, generous negative space at the top for a headline.
Colour variant test
Keep the product's shape, material, and label layout identical. Change only the body colour to deep forest green. Same lighting, same angle, same background.
Marketplace rules, translated to prompts
Most listing rejections come from prompting for a look instead of prompting for a specification. The left column is what the marketplace asks for; the right is the phrasing that gets you there. For the editing side of the same job — cleanup, batch consistency, variant sets — see the ecommerce photo editing guide.
| What the rule says | What to put in the prompt |
|---|---|
| Pure white main-image background | "pure white studio backdrop, RGB 255 255 255, no gradient or vignette" — asking for "a white background" often returns a soft grey studio sweep |
| Product fills most of the frame | "product centred and filling the frame with a small even margin" — then crop in yourself rather than re-generating |
| No text, logos, borders, or watermarks | "no added text, badges, borders, or graphics" — and generate with paid credits, or the watermark alone disqualifies it |
| High resolution so zoom works | generate at 2K or 4K for the main image only; drafts stay at 1K so you are not paying the multiplier on discards |
| Accurate representation of the item | keep the product region untouched and change only the environment — this is a prompting discipline, not a setting |
A workflow that costs almost nothing: draft at 1K on Shotari Basic until the composition is right, then re-run the one you kept on a paid model at 2K. You pay the higher rate once, on the image you actually ship, instead of on every attempt.
Try it with one packshot
Upload a single honest photo of your product and rebuild the studio around it. Free to start — no account needed for the first couple of generations.
Open the product photo editor