Photos for a 3D Product View: Which Set Do You Need?
Photos for a 3D product view: the shot list that decides whether the model reads right, how a one-photo view differs from a scan, plus the pre-send checks.
Photos for a 3D product view fail for reasons a photo count will not fix: the light moved between frames, the object turned on a turntable that shifted, or the surface was so polished the software had nothing to grip. Suppliers shoot eighty pictures, get a model with a melted handle, and blame the software.
The fix starts before the camera. Three routes produce a 3D view, and each one needs a different photo set, a different amount of time, and a different promise to the buyer. Pick the route first. The shot list follows from it, and the right photos for a 3D product view are the ones that match the route, not the largest set you can shoot.
This guide is for export-sales teams and suppliers who need a buyer to understand a product's shape without booking a flight to the factory. It is not about photogrammetry as a research method, and it is not about selling a 3D scanner.
Three routes, three photo sets
A 3D product view is any interface where a buyer can rotate a product on screen instead of flipping through static images. Under that label sit three different jobs, and mixing them up is the most common way a supplier wastes a week.
Route A: one clean photo, generated presentation model
A generated 3D product view is a rotatable model inferred from a single product photo. It is a presentation asset: it shows the shape and proportions a buyer would see from the front, and it opens in a browser without an app or plugin.
One photo is enough because the model is inferred, not reconstructed. That is also the limit. The back of the product, the underside, and the inside of a cabinet are plausible guesses, not evidence. Thin parts such as handles, cables, and table legs are where inference shows first. If the photo hides a leg behind the seat, the model may invent a shorter one or merge it into the frame.
This route fits catalog cards, WhatsApp links, trade-show QR pages, and any buyer handoff where the question is "what does it look like from another angle?" It does not fit a question about size, and it should never carry a dimension label. The numbers on a generated model were never measured.
Route B: twenty or more overlapping photos, photogrammetry
Photogrammetry reconstructs an object's shape from many overlapping photographs. Software finds the same points in frame after frame, solves where the camera stood, and builds a mesh from the matches. The photos are the data; the model is a visualisation of them.
The photo rules are unforgiving, and for photos for a 3D product view they are not about count:
Consecutive frames need at least 60% overlap. Less than that and the software cannot align the set, so the model fails to build or builds with holes.
You need more than one pass. A working pattern is one orbit near eye level, one at roughly 45 degrees, and one steep pass at 60 to 70 degrees. Three passes let the software see surfaces that one orbit hides.
Around a 12-inch turntable, 20 frames produce enough overlap for a model that usually builds. That is a starting point for a small object, not a universal number; bigger and more complex shapes need more.
Light must be diffuse and stationary. Hard shadows that move between frames get baked into the model as dark patches or confuse alignment.
The lens must not zoom between frames. Photogrammetry assumes one consistent scale across the set, and changing focal length breaks that assumption.
Some products fight this route no matter how many photos you take. Polished metal, glass, and high-gloss coatings reflect the room instead of showing surface detail, which is why scanning labs treat shiny objects as a separate problem with separate tricks. If your product is a mirror-finish tap or a glass vase, budget for retouching or choose another route.
Route C: no 3D at all, a measured spec diagram
A measured spec diagram is a product photo with dimension lines placed on the real outline, using numbers the supplier measured. It is the only one of the three routes that carries a size claim, because the numbers come from a tape or a caliper, not from a mesh.
When a buyer asks "will it fit through the door", "does the drawer clear the skirting", or "what is the seat height", shape is the wrong answer. A rotatable model invites the buyer to guess the scale. A dimensioned photo ends the guessing in one look. You can see the difference in a measured spec diagram demo: the product is the same, but the decision is made by the numbers on the image.
Decision matrix: which route fits your product
Route
What it proves
Minimum photo input
Best for
Fails when
A — generated
Shape impression from the front
1 clean photo, all edges visible
Catalog cards, chat links, QR pages, fast SKU coverage
Buyer needs the back, the underside, or an exact size
B — photogrammetry
Reconstructed surface geometry
20+ overlapping frames, 3 passes
Matte, high-value items where shape is the selling point
Product is shiny, transparent, or has thin moving parts
C — measured spec diagram
Real dimensions on the real product
1 sharp photo plus a tape measure
Fit questions, installation clearance, quoting
Buyer wants to inspect the surface from another angle
Read the matrix by the buyer's question, not by the product category. A sofa can need route C for the doorway and route A for the fabric close-up. A machined bracket can need route B for the casting shape and route C for the mounting holes. One SKU can use more than one route, as long as each asset makes only the claim it can support.
The shot list: photos for a 3D product view, by route
The table below is the part to hand to whoever holds the camera. It separates what route A needs from what route B needs, so nobody shoots a full orbit for a one-photo job.
Shot
Route A
Route B
Why it matters
What it prevents
Straight-on front, product filling most of the frame
Required
Required
The reference view a buyer compares against
A model that reads smaller or wider than the real item
Three-quarter view at about 45 degrees
Optional
Required
Shows depth and corner radius
A flat-looking model with no side plane
Side profile
Optional
Required
Confirms height-to-depth ratio
Stretched proportions
Top-down
No
Required
Captures the plan shape
A top face that is guessed or missing
Back or underside
No
Required if visible in use
Completes the surface set
Invented structure the buyer discovers later
Detail of hardware, seam, or texture
Optional
Required
Gives the mesh something to lock onto
Melted handles and smeared logos
Steep pass at 60 to 70 degrees
No
Required
Covers the surfaces the orbit misses
Holes on the top edge
Scale reference beside the product
No
Useful, not a substitute for measuring
Keeps the photo set honest about size
Someone quoting a size from the model
Photo hygiene matters as much as the list:
Matte beats glossy. If the product is glossy, shoot it with diffused light and accept that the scan may need retouching. Do not promise a clean scan of a mirror finish.
Lock the zoom. Set the lens to one focal length and leave it. Moving closer is fine; zooming is not.
Turn off the built-in flash. A flash that fires from a different position on each frame creates moving highlights, and moving highlights break alignment.
Keep the background plain for route A. A busy background gives the cutout more chances to include a chair leg that is not part of the product.
Shoot the product, not the catalogue. A photo that was already retouched for a brochure has lost the surface detail a scan needs.
If you are shooting a whole catalog rather than one SKU, the product photography style guide covers the fields that keep framing, camera height, and light consistent across a range. Consistency is what makes a set of 3D views look like one supplier rather than ten.
What the marketplace checks, and what it cannot
Amazon removed the 360-degree image experience from product detail pages on 20 January 2025. The replacement it promotes is a 3D model, and adding one costs nothing. Amazon accepts GLB or GLTF files, caps models at 200,000 triangles, and takes PNG or JPG textures. Sellers can upload an existing model through the Image manager or use the Create 3D Models feature in the Amazon Seller app to scan a product.
That is a file gate, not a quality gate. The platform can check triangle count, file integrity, and texture format. It cannot check whether the model still looks like the product you shipped. A model can pass every technical requirement and still show a cabinet with one door, a chair with three legs, or a handle that melted into the frame.
The 360 image is gone, but the older habit is not. Suppliers still ask for "a spin" and expect it to work on a marketplace detail page. On your own site, a 360 sequence is still a valid asset. It is not a 3D model, and it does not answer the same question. If you are choosing between the two formats, the trade-offs are laid out in 3D product viewer vs 360 photos.
One more boundary worth writing into the brief: a generated 3D view is not engineering geometry. It does not certify wall thickness, thread pitch, load capacity, or hidden structure. If the buyer needs those, they belong in the spec table and on the drawing, not in the model.
Six checks before you generate or send
Run these on the model, not on the source photo. The photo can be perfect and the output still wrong.
Silhouette check. Compare the model's outline against the front, side, and top photos at the same viewing angle. If the outline drifts, stop.
Thin-part check. Look at handles, legs, cables, and trim. Are they intact, or fused into the body?
Face-count check. Count the visible faces against the real product. A missing back panel is easy to miss on a screen and obvious to a buyer.
Texture placement check. Confirm the grain, logo, or colour is on the correct surface, not wrapped across a seam.
File check. Confirm GLB or GLTF, under the platform's triangle cap, textures in an accepted format, and a file size that loads on a mid-range phone.
Claim check. Make sure no size number appears on or beside the 3D view unless it came from a measured drawing. Keep the dimensions in the spec diagram.
The last check is the one that protects the order. A 3D view sells shape; a spec diagram sells size. Confusing the two is how a supplier wins the click and loses the order.
Where the 3D view sits in the handoff
A rotatable model is most useful when it is attached to the same record as the specs, not sent as a loose file. A buyer who opens a link, turns the product, and reads the dimensions in the same view does not have to ask which version is current. A rotatable 3D product model belongs on that page as the main view, with the measured numbers in the table beside it.
That is also the practical limit of the asset. A 3D view helps a buyer understand shape, not certify a specification. The supplier still supplies the numbers, and the numbers still need a tape measure behind them.
FAQ
How many photos do you need for a 3D product view?
There is no universal number. A generated one-photo view needs one clean, sharp photo with every edge visible. A photogrammetry scan needs a full overlapping orbit, and 20 frames around a 12-inch turntable is a working baseline for a small object. Larger or more complex products need more frames and more passes.
Can you make a 3D product model from one photo?
Yes, for presentation. A single photo can produce a rotatable model that shows the front shape and proportions in a browser. The parts the photo does not show are inferred, so the back, underside, and interior should not be treated as evidence. One-photo models are for catalog cards and buyer handoffs, not for verifying structure.
Is a 3D product view accurate enough to quote dimensions from?
No. A 3D view built from photos or inferred from one image has no measurement authority, even when the proportions look right. Quote from a measured drawing or a dimensioned photo, and keep the 3D view for shape. If the buyer's question is fit, the measured asset is the one to send.
What file format do marketplaces accept for a 3D product view?
Amazon accepts GLB or GLTF, caps models at 200,000 triangles, and takes PNG or JPG textures. Other marketplaces publish their own requirements, so check the current help page before you export. A file that passes one platform's format check may still fail another's size or texture rule.
Why do shiny or glass products fail in 3D scanning?
Photogrammetry needs stable surface detail to match points across frames. Polished metal and glass reflect the room, so the same point looks different in every photo and the software cannot lock on. The usual workarounds are diffused light, anti-reflective treatment, or a different capture method entirely.
