Image to 3D model: how a photo becomes a 3D asset

Quick answer: Image to 3D model software takes ordinary product photos and turns them into a rotatable 3D asset you can put on a product page. There are three practical routes, and they differ on price, time, and how much cleanup lands on your desk:

• AI generators take one or a few photos and output a mesh in minutes. Subscriptions commonly land between $39 and $199 per month (Shopify’s 2026 tool roundup lists live examples at $39, $69.90, $99 and $199), with free tiers in the mix. The mesh usually needs cleanup before it goes near a product page.
• Human modelers, freelancers or studios, model your product from reference photos. Shopify’s own blog puts simple product models at $40 to $200 and complex ones past $1,000; a 2026 cost calculator puts most base models at $140 to $1,500. Days to weeks, per model.
• Managed pipelines, including Peka AR, combine AI generation with a human artist pass and hand back a web-ready file in hours.

Whichever route you pick, the file your storefront wants is GLB for the web, with USDZ alongside so iPhones can open it in AR. And because a photo only shows one side of the product, the unseen sides get inferred: more reference angles, closer match to the real thing.

That is the whole answer in short. Below is what each route really produces, what cleanup actually means, and how to choose.

What image to 3D model software does with a photo

Every photo-to-3D tool is filling in the parts of the object the camera never saw. Two techniques dominate.

Photogrammetry takes many photos from overlapping angles and triangulates real geometry from them. It is accurate when you shoot enough frames, which is why scanning apps ask you to walk around the object.

AI generation feeds the image through a trained model that has seen hundreds of thousands of objects and predicts a plausible mesh. One photo is enough. It is also why the back of your product is a guess: the model renders what objects like yours usually look like, not what yours looks like.

A third path, manual modeling, skips reconstruction entirely. An artist opens Blender or Maya, uses your photos as reference, and builds the geometry on purpose. Slowest, most controllable.

All three end at the same place: a mesh, some textures, and a long gap between that file and something you would put in front of a paying shopper.

The three routes, with real numbers

Route	Typical cost	Time per model	Who fixes problems
AI generator subscription	~$39–199/mo, free tiers exist	minutes	you
Freelancer or studio	$40–200 simple (Shopify, 2020); $140–1,500 base (Eyedex, 2026)	days to weeks	usually already done
Managed AI + artist pipeline	flat subscription credits	hours	the service

AI generators: fast and cheap, output is raw

Tooling here has gotten good enough that the first pass appears in under ten minutes. What you get back is a draft: geometry that may fuse nearby parts, textures that stretch across surfaces the photo never showed, and file weights nobody would ship to a phone. Every serious comparison of these tools says the same thing in different words: expect a cleanup pass.

If you know Blender, that trade is fine. Generating drafts for internal review is exactly what these tools are for. The live pipeline demo on our homepage runs a similar draft-first flow. If you do not know Blender, a raw mesh is where the project stops.

Freelancers and studios: finished, but slow and quote-shaped

Post the product, collect bids, compare portfolios, manage revisions across a timezone. Output quality is real when you find the right person, and the price follows complexity: simple objects at the low end, jewelry and machinery at the top. Shopify’s own guidance to merchants puts basic models at $40 to 200 with complex work past $1,000 (Shopify, 2020); a 2026 market calculator puts typical base models at $140 to 1,500 before variants and animation (Eyedex, 2026).

The hidden cost is coordination. One model is a project. Twenty models is a vendor you now manage.

Managed pipelines: AI draft, human finish, hours not weeks

This is the category Peka AR sits in, so treat the specifics as our pitch, not neutral research. Our AI pipeline returns a draft in around 5 to 10 minutes, clearly labeled as AI throughout the app, and the Artist tier puts a 3D artist on every model before it ships: likeness checked, textures rebuilt where the guess was wrong, geometry optimized for the web, delivered in hours. One price covers generation, review, hosting, and the embed. See pricing.

Other vendors run managed services too. The tell is whether they publish a price and a turnaround, and whether a human touches the model before you do.

What “cleanup” means when people say AI output needs it

Four separate problems hide in that word:

1. Geometry. Non-manifold edges, wasted polygons, parts that should be separate fused into one blob. Fixing it means retopology, which is the job people mean when they say cleanup.
2. Textures. Occluded surfaces (the back, the underside) get smeared or hallucinated. Logos and label text come out warped more often than not. Shopify’s partner guidance calls out logos and labels as needing special attention and recommends keeping the whole model near 4 MB with mobile textures capped at 2048×2048 (Shopify partner docs).
3. Likeness. The model looks like a chair. Does it look like your chair, in the right color and material? A shopper who rotates a slightly-off version of the product is worse off than one looking at a photo.
4. Web budget. Untouched AI exports are heavy. On a mid-range phone over cellular, a 40 MB model is a poster frame that never finishes loading.

An artist pass is just these four fixes done by someone whose job they are.

GLB and USDZ, the two formats that matter

• GLB is binary glTF: geometry, textures, and materials packed in one file. It is what web viewers like Google’s model-viewer render, and what every self-serve viewer expects.
• USDZ is Apple’s AR container. It is what iOS Quick Look opens when a shopper taps “view in your space.”

You need both if you want AR on both major phone platforms. Shopify accepts either and handles the pairing for you: per Shopify’s docs, files over 15 MB are automatically optimized, and uploads are converted so both a GLB and a USDZ version exist for Android and iOS (Shopify, product media types).

Getting the model onto the storefront

Two paths work, and the choice is mostly about your theme:

1. Shopify’s built-in 3D product media. Upload the GLB into the product’s media, use a theme that supports 3D (all Online Store 2.0 Shopify themes do), and Shopify serves the viewer plus the AR conversion. Full walkthrough: how to add a 3D model to your Shopify product page.
2. An iframe embed. One line of HTML, works on any theme or platform (Shopify, WooCommerce, custom), and the viewer plus AR button arrive as a package. This is how our own delivery works; the how-it-works section shows the pipeline behind it.

Is a 3D product page worth it? What the research reports

Worth citing carefully, because these are third-party studies, not our numbers:

• Harvard Business Review reported that shoppers using AR were 19.8% more likely to purchase (HBR, March 2022).
• An Alter Agents study with Snap and Publicis Media found 80% of AR shoppers felt more confident buying after using it (Alter Agents, July 2022).
• Online return rates sit near 30%, with “product looked different” among the top reasons (Total Retail, reporting industry data).

The mechanism is not mysterious. Rotation and room-scale AR replace the guesswork that flat photos leave behind, which is the guessing that returns are made of. The benefits section of our landing page lays out where this applies to a real storefront and where it does not.

How to choose, in one pass

• Experimenting with one or two products, comfortable in 3D software: an AI generator subscription. Cheapest way to learn what raw output looks like.
• A handful of hero SKUs, deadline exists, budget exists: a freelancer or studio. You are buying craft and management at the same time.
• A catalog to get live, repeatedly, without running a vendor: a managed pipeline. You are buying turnaround and someone else’s QA.

Whichever you choose, the photo prep is the same: plain background, even light, a dozen angles including the back and underside, no props covering the product. The better the reference, the less anyone has to guess.

Pick three products you would love shoppers to rotate, pick your route above, and start there. Three finished models on real product pages teach you more than a folder of thirty drafts.