Image to 3D Model: How to Convert Photos to 3D in 2026

Turning a regular photo into a 3D model used to require professional 3D artists and weeks of work. In 2026, you have two solid paths: photogrammetry (taking many photos from different angles) and AI-based generation (feeding a single image to a neural network). Both have gotten dramatically better in the past year.

This tutorial covers both approaches with specific tools, step-by-step workflows, and honest quality assessments.

The Two Approaches

Photogrammetry: Many Photos to One 3D Model

You photograph an object from 30-200+ angles. Software analyzes the overlapping images, identifies matching features, and reconstructs the 3D geometry. This is the more established technique and produces higher-quality results.

Best for: Physical products, real-world objects, anything where you need accurate geometry and textures.

Requires: 30-200 photos of the same object, taken from every angle. Good lighting. A turntable helps.

AI Single-Image 3D: One Photo to One 3D Model

Neural networks (trained on millions of 3D objects) estimate the 3D shape from a single 2D image. This sounds like magic, and the results are improving fast, but they’re still below photogrammetry quality.

Best for: Quick prototypes, concept visualization, social media content, cases where photogrammetry isn’t practical.

Requires: One clear photo. That’s it.

Method 1: Photogrammetry (Step by Step)

What You Need

• A camera or smartphone (iPhone 12+ or any phone with LiDAR is ideal, but any camera works)
• A turntable ($20-$50 on Amazon) or the ability to walk around the object
• Consistent, diffused lighting (avoid harsh shadows and reflections)
• Photogrammetry software (free and paid options below)

Step 1: Capture Your Photos

This is where quality is made or lost.

For small objects (products, sculptures, shoes):

1. Place the object on a turntable with a plain, non-reflective background
2. Set up 2-3 diffused light sources (or use a lightbox)
3. Lock your camera’s exposure and focus. Don’t let auto-exposure shift between shots.
4. Take 36-72 photos at one elevation (every 5-10 degrees of rotation)
5. Change camera height and repeat. Aim for 3 elevation rings: low, eye-level, high
6. Take 10-20 close-up detail shots of complex areas
7. Total: 120-200+ photos for best results, minimum 40-60 for basic models

For large objects (furniture, cars, buildings):

1. Walk around the object in a circle, taking a photo every 2-3 steps
2. Repeat at different heights if possible
3. Get close-ups of detailed areas
4. Total: 60-200 photos

Critical rules:

• No blurry shots. Delete any that aren’t sharp.
• Overlap between consecutive photos should be 60-80%. Each photo should share most of its content with the previous one.
• Don’t move the object between shots.
• Avoid transparent, reflective, or very shiny objects. Photogrammetry can’t handle mirrors, glass, or chrome (the software gets confused by reflections).

Step 2: Process with Software

Meshroom (Free, open source)

The best free option. Runs on Windows and Linux (no Mac).

1. Download from alicevision.org
2. Drag and drop your photos into the Meshroom window
3. Click “Start” and wait. Processing 100 photos takes 30 minutes to 2 hours depending on your GPU.
4. The output is a textured 3D mesh (OBJ format with texture maps)

Polycam ($7.99/month)

Mobile-first photogrammetry. Works directly on iPhone and Android.

1. Open Polycam, select “Photo Mode”
2. Walk around the object while the app guides you on coverage
3. Upload to Polycam’s cloud for processing (takes 5-15 minutes)
4. Download as GLB, OBJ, FBX, or USDZ

Polycam’s advantage: the app tells you in real-time if you’ve captured enough angles. The quality is slightly below Meshroom on the same photo set, but the workflow is much easier.

RealityCapture ($10/month or pay-per-model)

The professional standard. Best quality, fastest processing, steepest learning curve.

1. Import photos
2. Align photos (automatic)
3. Build mesh (multiple quality levels)
4. Texture the mesh
5. Export in your format of choice

RealityCapture handles massive datasets (1000+ images) that choke other tools. If you’re doing this professionally, it’s the one to learn.

Agisoft Metashape ($179 one-time)

The academic and surveying standard. Comparable quality to RealityCapture, slower but more configurable.

Step 3: Clean Up the Model

Raw photogrammetry output always needs cleanup:

• Remove stray geometry (parts of the background, turntable fragments)
• Fill holes where the software couldn’t reconstruct (usually the bottom of the object)
• Reduce polygon count for web use (100K-500K polygons is typical for photogrammetry output, you want 10K-50K for web)
• Fix textures if there are seams or blurry spots

Free tools for cleanup: Blender (powerful but complex), MeshLab (simpler, mesh-focused).

Method 2: AI Single-Image 3D

These tools use neural networks to estimate 3D geometry from one photo. The technology is based on research like NeRF (Neural Radiance Fields) and 3D Gaussian Splatting.

Meshy

The most popular AI-to-3D platform right now.

1. Go to meshy.ai
2. Upload one clear photo of your object
3. Select “Image to 3D”
4. Wait 1-3 minutes
5. Download as GLB, OBJ, FBX, or STL

Quality: Impressive for organic shapes (animals, characters, food items). Geometry is approximate but convincing from a distance. Textures are painted on based on the input image, so the back of the model is AI-guessed.

Cost: 200 free credits on signup. Pro at $20/month for 500 credits.

Luma AI (Genie)

Luma started with NeRF captures and expanded to AI 3D generation.

1. Use the Luma Genie API or web interface
2. Upload an image or provide a text prompt
3. Choose quality level (draft, standard, high)
4. Download the result

Quality: Better textures than Meshy for some categories, especially architectural objects and everyday items. Worse on fine geometric details. The “back” of objects is often simplified.

Cost: Free tier available. Pro from $9.99/month.

CSM (Common Sense Machines)

More developer-focused. Offers an API for automated 3D generation.

1. API call with image URL
2. Receive 3D model back in GLB format
3. Process time: 30 seconds to 2 minutes

Quality: Competitive with Meshy. Stronger on hard-surface objects (electronics, furniture). Weaker on organic forms.

Cost: API pricing, roughly $0.10-0.50 per model.

Quality Comparison

Let’s be honest about where these tools stand:

Quality ratings based on geometric accuracy, texture fidelity, and production-readiness for e-commerce use.

Method	Geometry Accuracy	Texture Quality	Back of Object	Production Ready?
Photogrammetry (pro)	Excellent	Excellent	Real (captured)	Yes
Photogrammetry (free)	Good	Good	Real (captured)	Yes, with cleanup
AI single-image	Approximate	Painted/guessed	AI-generated	For prototyping only

Bottom line: Photogrammetry produces models you can put on a product page today. AI single-image produces models good enough for concept work and social media, but not for serious e-commerce where customers will inspect the model from every angle.

Hosting and Displaying Your 3D Model

Once you have a 3D model, you need somewhere to put it. A GLB file sitting on your server needs a viewer to become interactive.

Self-Hosted Viewers

model-viewer (Google): Free, open-source web component. Add <model-viewer src="model.glb" ar> to your HTML and you’ve got a 3D viewer with AR support on mobile. Requires hosting the GLB file yourself.

Three.js: Full 3D engine for custom viewers. More control, more code. Takes 2-5 days to build a production-quality viewer from scratch.

CDN-Hosted Viewers

Sirv: Upload GLB/glTF/USDZ files and embed with one line of code:

<div class="Sirv" data-src="https://your-account.sirv.com/model.glb"></div>
<script src="https://scripts.sirv.com/sirvjs/v3/sirv.js"></script>

Sirv’s 3D viewer includes AR via WebXR (Android) and iOS Quick Look (USDZ auto-generated), material customization, and the model is served via CDN from 24 global PoPs. If you’re already using Sirv for product images and 360 spins, adding 3D models keeps everything in one place.

Sketchfab: The YouTube of 3D. Upload models, embed on your site, share publicly. Free tier with watermark, Pro from $15/month.

File Format Quick Reference

Format	Use Case	AR Support	File Size
GLB	Web delivery (binary, single file)	Android WebXR	Smallest
glTF	Web delivery (JSON + separate assets)	Android WebXR	Small
USDZ	iOS AR Quick Look	iOS only	Medium
OBJ	Editing, 3D printing	No	Large
FBX	Game engines, animation	No	Large
STL	3D printing	No	Large (no textures)

For web use, GLB is the standard. It bundles geometry, textures, and materials into one file. Most photogrammetry and AI tools can export directly to GLB.

Your Next Move

If you have the product in hand: Try photogrammetry first. Download Meshroom (free), take 60+ photos with your phone, and process them. The quality will surprise you.

If you just have a photo: Try Meshy’s free tier. Upload a clean product shot on white background and see what comes back. The result will be rough, but it gives you a preview of what AI 3D is capable of today.

If you’re building a product page: Focus on photogrammetry for accurate models, host them on a CDN with a built-in viewer, and enable AR so mobile users can see your product in their space. That combination of accurate geometry and easy deployment is what actually converts shoppers.