Should I put the real product into an AI scene instead?

For jewelry with stones, often yes. Photograph the piece in a light tent, cut it out, and place it into an AI-made scene or onto a body, so the stone stays real and only its surroundings are generated. Match the light direction and add a contact shadow so it sits in the scene.

Does image size matter for small products?

Yes, because a clasp or a stone covers few pixels. Google prices Nano Banana Pro at $0.134 for an image up to 2K and $0.24 at 4K (September 2026). Rendering large and cropping in can keep small parts cleaner than rendering small, as long as you export within the marketplace’s size limits.

Can shoppers try on shoes virtually?

On Google, yes: Google added shoes to its shopping try-on on October 8, 2025, working from a full-length photo of the shopper. Google’s help page says accessories are not supported. See virtual try-on vs AI model photos.

How to make AI photos of shoes, jewelry and accessories

Checked September 24, 20266 min read

Short answer

Shoes, jewelry and accessories are hard for image models because they are small, shiny and symmetrical. Metal reflects a room the model invents, stones lose their sparkle or change size, pairs stop matching, and laces and stitching get redrawn. Shoot sharp references from every side, describe soft, surrounding light, keep marketplace main images plain (Amazon wants one left-facing shoe, and jewelry without a model), and check every detail at full size.

What makes these products hard

Product	What AI gets wrong	Why
Polished metal: rings, buckles, watch cases	Reflections of a room or sky that is not there, warped edges, the wrong metal color	Shiny metal shows its surroundings, and the model invents the surroundings
Gemstones and diamonds	Sparkle turns flat and white, facets blur, stones grow	A stone’s look comes from how its cut handles light, which a model imitates instead of calculating
Small parts: clasps, chain links, eyelets	Links merge, clasps change type, counts drift	Small parts cover few pixels, so there is little detail to hold on to
Pairs of shoes and earrings	Left and right stop mirroring each other; one gets extra details	Nothing in the model forces the two to match
Laces and stitching	Lace paths that could not exist, stitch counts that change, stitching that crosses panels	Repeating patterns are hard to keep consistent
Logos and stamps	Letters redrawn on buckles, soles, tags and clasps	The same weakness as any text in an image
Bags	Strap anchors move, handles change length, the lining changes color	Unless you send them, the model has never seen the inside or the strap drop

The fixes for text and logos are in why AI changes your product.

References to shoot first

Shoes: each shoe from the outside, inside, front, back, top and sole, the pair together, and close-ups of laces, stitching and any logo
Jewelry: the piece from the front and side, the clasp or back, each stone close up, and one photo on a hand, ear or neck for scale
Bags: front, back, side, base, inside, the strap anchors, and the bag worn so the strap drop shows
Watches: the dial straight on, the case side with the crown, the clasp and the caseback
One photo with a ruler or a familiar object, so you can check scale later
Every piece cleaned first, because dust and fingerprints show at macro size

For the phone setup, see how to photograph your product for AI.

Light metal and stones the way a studio does

Jewelry photographers soften and surround the light, because polished metal reflects whatever is around it. In a Nikon USA article, commercial photographer Jody Dole uses a light tent to diffuse the light and keep results consistent, adds a second light for dimension, and shoots on macro lenses at 1:1 (life size), with apertures from f/8 to f/32 in his examples. He also keeps canned air on set, because dust shows.

Write that setup into the prompt as light acting on the object: "the ring sits inside a white diffusion tent, a large soft light from the upper left wraps the band, a thin dark card on the right draws one clean dark line along the polished edge, and the stone catches small bright points of light." More light words are in lighting terms for AI prompts.

For stones, GIA, the Gemological Institute of America, describes three effects: brightness (white light reflected back), fire (light split into spectral colors) and scintillation (the sparkle and the pattern of light and dark when the stone moves). Ask for the ones the real stone shows, and compare the render with a photo of the real stone. Watch for added fire. A stone with little of it should not throw rainbow flashes in the render.

On-body shots and how they fail

Shot	How it fails	How to brief it
Ring on a hand	Extra or merged fingers, the ring floating above the skin, the wrong finger	Say which hand and finger, how many fingers show, and that the band sits at the base of the finger with a slight shadow
Earrings on an ear	Hanging from the wrong point, left and right different, hair covering them	Name the piercing point, show one ear in profile, hair tucked behind
Necklace on a neck	Chain length changes, the pendant grows, the clasp appears at the front	Give the chain length in inches or centimeters and where it lands: collarbone, sternum
Shoes on feet	Feet too big for the shoe, impossible lacing, soles floating	Give the shoe size and a real model height, and ask for contact shadows under the soles
Bag on a shoulder	Strap length changes, the bag’s scale is off against the body	Give the strap drop, the bag’s dimensions and the model’s height

If a hand keeps failing, price a real one; soona lists hand models from $79 an hour and foot models from $99 an hour (September 2026). More on grips and fingers in how to get hands right in AI product photos.

Marketplace rules for main images

Amazon US rule, as of September 2026	What it means for AI photos
Footwear main image: a single shoe, facing left, at a 45-degree angle	Make the left-facing single shoe on pure white; keep pairs and on-foot shots for secondary images
Jewelry main image: pure white background, the actual product, no mannequin or model, no text or logos	No on-body AI shot as the main image
Jewelry images: at least 1,000 pixels on one side for zoom, no more than 2,100 pixels on the longest side, and the product filling at least 80 percent of the frame	Render large, crop tight, then export at 2,100 pixels or less. Amazon’s general image guide allows up to 10,000 pixels and asks for 85 percent fill, so check what your upload accepts
Jewelry images: only what the customer receives	No extra pieces or props in any image
Clothing accessories and multipacks: main image flat, without a model	Belts, scarves and hats go flat

The FTC’s Jewelry Guides say it is deceptive to misrepresent a jewelry product’s size, color, cut or other material features, and they note that showing a diamond or other gemstone larger than its real size may mislead unless the true size is disclosed. An AI render that enlarges or brightens a stone risks exactly that. Amazon’s jewelry page tells sellers to follow the FTC’s guidelines for the jewelry industry.

Checks before anything goes live

Metal color matches: yellow, white or rose gold, silver, brass, gunmetal
Reflections show a plain studio, with no invented room, sky or window
Stone size, count, color and cut match the real piece, with no added fire or sparkle
Chain links, clasp type and setting prongs match
Pairs mirror each other: same height, same details on the outside of each shoe or earring
The lace path is possible and matches the real lacing, with the right number of eyelets
Stitching follows the real panels, in the right color
Logos on buckles, soles, tags and clasps are spelled right
Scale looks right against the hand, ear, foot or body
Amazon main images: one left-facing shoe, no model for jewelry, pure white

Then run the full AI image QA checklist.

Questions people also ask

Should I put the real product into an AI scene instead?: For jewelry with stones, often yes. Photograph the piece in a light tent, cut it out, and place it into an AI-made scene or onto a body, so the stone stays real and only its surroundings are generated. Match the light direction and add a contact shadow so it sits in the scene.
Does image size matter for small products?: Yes, because a clasp or a stone covers few pixels. Google prices Nano Banana Pro at $0.134 for an image up to 2K and $0.24 at 4K (September 2026). Rendering large and cropping in can keep small parts cleaner than rendering small, as long as you export within the marketplace’s size limits.
Can shoppers try on shoes virtually?: On Google, yes: Google added shoes to its shopping try-on on October 8, 2025, working from a full-length photo of the shopper. Google’s help page says accessories are not supported. See virtual try-on vs AI model photos.

Where Overs fits

In Overs, Nano Banana Pro (about $0.134 a photo) is the model for fine detail and matching references, and "Finish the keepers" remakes approved photos on a better model or quality. On Pro and Team plans, a reviewer checks each photo against the references before you see it, and can retry on its own up to a limit you set.

Free for 40 photos a month. The AI that makes the photos is billed separately, on your own key, with no markup from Overs.

Start free How it works

Sources

Checked on September 24, 2026. Prices, specs and rules change; follow the links for the current versions.