Sourcing Models
Introduction
When building complex 3D scenes for visual investigations, you don't always need to model everything from scratch or reinvent the wheel. Sourcing existing 3D models can significantly accelerate your workflow, providing ready-made assets that can be integrated into your reconstruction work.
This page will cover some key considerations when sourcing models:
- Quality assessment - Evaluating origin, polygon count, texture resolution, and accuracy
- Licensing and attribution - Understanding usage rights and proper crediting
- Format compatibility - Ensuring models work with your Blender workflow
- Contextual appropriateness - Finding models that match your investigation's location, time period, and scale
This page mainly focuses on Sketchfab, one of the largest platforms for finding, previewing, and downloading 3D models. Sketchfab offers several key advantages for visual investigators: Extensive Library with Quality Models, Inspect Before Downloading, Clear Licensing Information, often Free downloads with Multiple Formats.
Sketchfab Platform
Access to millions of 3D models with real-time preview, licensing information, and direct download capabilities.
Licensing Clarity
Clear Creative Commons licensing options that define how you can use models in your investigation work.
Quality Filters
Search and filter tools to find models by polygon count, animation status, rigging, and file format.
Sources for 3D Models
While this page focuses primarily on Sketchfab, you should be aware that there are a variety of platforms and resources for sourcing 3D models, more than I can or should mention here. I still want to highlight some, each has distinct advantages depending on your needs.
| Platform | Type | Best For | License/Cost | Notes |
|---|---|---|---|---|
| Sketchfab | Community/Commercial | General purpose, broad library | Free & Paid, Various CC licenses | Largest library, real-time preview |
| Polycam Gallery | User Photogrammetry | Recent location-specific scans | Free tier available | Mobile captures, quality varies |
| Luma AI Gallery | NeRF Captures | Clean architectural details | Free tier available | Newer technology, smaller library |
| Smithsonian Open Access | Cultural Institution | Museum artifacts, cultural objects | CC0 (Public Domain) | Verified accuracy, high quality |
| OpenHeritage3D / CyArk | Cultural Heritage | Archaeological sites, monuments | Various (often CC) | Historical documentation included |
| CG Trader / TurboSquid | Commercial | Specific vehicles, technical objects | Paid (commercial license) | Professional quality, rigged/animated |
| BlenderKit | Blender Add-on | Direct in-software access | Free & Paid | Browse without leaving Blender |
| Google Earth 3D | Tech Platform | Global coverage | ⚠️ Extraction violates ToS | Use only as reference, not extraction |
Google Earth / Google Maps
Important Legal Note: Extracting 3D models from Google Earth violates Google's Terms of Service and is generally not permitted (or a 'grey area' at least) for most uses.
The Crowdsourced Era (2006-2013) For several years, 3D buildings in Google Earth and Google Maps were largely crowdsourced, built by members of the general public. Before Google had automatic scanning technology, they relied on volunteers to build the world by hand using two main tools:
- Google SketchUp: Google bought SketchUp in 2006 specifically for this purpose, releasing a free version so anyone could model their local buildings.
- Google Building Maker: A simpler browser-based tool launched in 2009 that showed aerial photos from different angles, allowing users to "connect the dots" to create rough 3D shapes.
Google shut down the program in 2013 and purged most user-submitted models from the main map. However, millions of these hand-made models are still hosted and downloadable on the 3D Warehouse, which is owned by Trimble (who bought SketchUp from Google in 2012). Some locations in Google Earth still display these legacy models.
The Rise of Aerial Photogrammetry (2013-Present) Google transitioned to automated 3D capture using aerial photogrammetry. This produces the "melted" appearance sometimes visible in Google Maps today.
The Technique: Oblique Aerial Photography Standard satellite imagery captures photos looking straight down (called the nadir angle), which is great for maps but terrible for 3D—you only see rooftops, not building facades. To solve this, Google contracts aerial imagery companies to fly over major cities in tight grid patterns. These planes carry specialized camera rigs that capture images in multiple directions simultaneously:
- v1 camera looking straight down (nadir)
- 4+ cameras angled at approximately 45 degrees (north, south, east, west)
These 45-degree "oblique images" capture building sides, allowing photogrammetry software to reconstruct full 3D geometry of structures, vegetation, and ground features. The result covers large areas with reasonable quality, though less sharp than hand-made models.
Neither of these models can be downloaded directly. However, there are options to get them into Blender.
Renderdoc & Ninjaripper
Both RenderDoc and Ninja Ripper are tools that intercept the communication between a program (like a game or Google Earth) and your graphics card.
You can think of them as "wiretaps" for your GPU. When a game tells your graphics card, "Instance an object here with this texture and this animation," these programs record that instruction and save the 3D shape, the texture and the animation to your hard drive.
- RenderDoc
- Ninja Ripper
RenderDoc is a professional, open-source graphics debugger. It is the industry standard tool used by game developers (like Ubisoft or Epic Games) to fix bugs in their own graphics code.
- How it works: It captures a single frame of a running application. It freezes time and lets you inspect exactly how that one frame was constructed, step-by-step.
- What it sees: It sees every single "Draw Call." You can scroll through a list and see: "First the sky was drawn, then the terrain, then the character, then the shadow..."
- The "Ripping" Aspect: Because RenderDoc has all this data, people found a way to "export" the 3D meshes and textures from the capture file into formats like CSV or OBJ that Blender can read.
- Your Context (Maps): This is the primary tool used to extract 3D data from Google Maps. Users inject RenderDoc into their web browser, capture a frame of the 3D city, and then use a specific Blender add-on to reconstruct that city block in Blender.
Ninja Ripper is a tool originally designed to extract assets from games. It is not meant for debugging; it is meant for "ripping." But since the principle is the same, we can also apply it to Google Maps/Earth.
- How it works: Instead of analyzing a frame, it aggressively grabs every 3D object and texture currently loaded in the video memory and dumps them into a folder on your computer.
- The Output: You end up with thousands of files named
mesh_001.rip,mesh_002.rip, etc. It is very unorganized and messy.
Consider these options either a last resort, or a reference point.
Effective Search Strategies
When searching for 3D models online, there are some straightforward and common sense strategies:
Be Specific
✗ "car"
✓ "2015 Toyota Hilux"
✓ "BMP-2 infantry fighting vehicle"
✓ "Damascus Umayyad Mosque"
Add Technical Terms
"photogrammetry" - Real-world scans
"scan" - Laser or photogrammetry captured
"pbr" - Physically-based rendering materials
"3D model" - Ensures 3D results (not images)
"3d" - Broader than "3D model"
Specify File Formats
"Car model X fbx"
"Landmark Y obj"
"building Z glb"
Combine Terms
"AK-47 pbr scan"
"concrete barrier 3d model obj"
"soviet apartment photogrammetry fbx"
Common Issues
Scale and Orientation Issues
Models often import at incorrect scales or orientations. This happens because:
- Different software uses different unit systems
- Artists work at arbitrary scales
- Blender uses Z-up, while many programs use Y-up
Always check scale and orientation immediately after import. Use reference objects (2m cube for human scale) and verify against real-world dimensions.
Overly Dense Topology
Photogrammetry scans and high-detail models can contain millions of polygons, causing performance issues.
Check polygon count in Blender's statistics. Use the Decimate modifier to reduce density for background elements. Consider creating lower-detail versions for distant objects.
Quality Variation Across Sources
When sourcing for models you'll encounter a variety of scenarios. Models come from different sources with varying quality:
Photogrammetry Scans
- High accuracy to real-world objects
- Often messy topology with holes or artifacts
- Very high polygon counts
- Excellent textures from real photos
- May have incomplete coverage (e.g., reflective surfaces don't transform well)
Artist-Created Models
- Usually clean topology but geometry may be stylized
- Can be too detailed or too simplified
- May not match real-world proportions exactly
- Textures can be artistic interpretation
Game-Ready Models
- Optimized polygon count
- Good for performance
- May lack fine detail
CAD Conversions
- Technically accurate dimensions
- Often no textures or materials
- Can have extremely complex geometry
- May need significant cleanup
Making the Right Choice
Ask yourself:
-
How much detail do I need?
- Background element: Low-poly, simple textures
- Mid-ground context: Medium detail
- Focus of investigation: Highest accuracy possible
-
What's my priority: speed or accuracy?
- Need it fast: Use whatever works, note limitations
- Need it accurate: Take time to find photogrammetry or verify measurements
-
Can I improve what's available?
- An incomplete photogrammetry scan of the exact building you need is often better than a complete generic building
- A messy scan can be cleaned up
- A low-poly model can be detailed if the proportions are correct
Even an incomplete or messy scan can be a better starting point than modeling from scratch. You can always fill holes, clean up topology, and improve textures as needed. Do assess how much time you think you'll spend and where. There is a delicate balance between how much detail you need, how much time you'd spend on cleaning a model vs. how much time you'd spend to build something from scratch.
Sketchfab Overview
Sketchfab is a web-based platform for publishing, sharing, and discovering 3D, VR, and AR content. For visual investigators, it offers several key advantages:
Platform Features
- Real-time 3D viewer - Inspect models directly in your browser before downloading
- Extensive library - Millions of models across every category imaginable
- Advanced search - Filter by license, polygon count, animated/rigged status, and more
- Model information - View topology, texture maps, and technical specifications
- Direct downloads - Export models in multiple formats including FBX, OBJ, and glTF
Search Capabilities
When searching Sketchfab for models relevant to your investigation, use these strategies:
Location-based searches:
- Search by city, landmark, or geographic location names
- Use terms like "photogrammetry" or "scan" combined with location names
- Look for heritage documentation and cultural preservation projects
Quality indicators:
- Filter by "Downloadable" to show only models you can export
- Sort by "Most Viewed" or "Most Liked" for community-vetted quality
- Check polygon count before downloading (aim for appropriate detail levels)
Understanding Licensing
One of the most important aspects of sourcing models is understanding licensing. Sketchfab models come with various Creative Commons licenses that determine how you can use them.
Common License Types
CC BY
Attribution
- ✅ Commercial use allowed
- ✅ Remix and modify allowed
- ⚠️ Must credit the creator
Good for most work
CC BY-SA
Attribution-ShareAlike
- ✅ Commercial use allowed
- ✅ Remix and modify allowed
- ⚠️ Must credit the creator
- ⚠️ Derivatives use same license
CC BY-ND
Attribution-NoDerivs
- ✅ Commercial use allowed
- ❌ Cannot modify the model
- ⚠️ Must credit the creator
CC BY-NC
Attribution-NonCommercial
- ❌ No commercial use
- ✅ Remix and modify allowed
- ⚠️ Must credit the creator
May limit some applications
CC0
Public Domain
- ✅ No attribution required
- ✅ Complete freedom to use
- ✅ Complete freedom to modify
Ideal but less common
Attribution Best Practices
When using sourced models in your investigation, proper attribution is both legally required and professionally important.
- In documentation: Include a "3D Assets" section listing all sourced models
- In reports: Credit creators in footnotes or appendices
- Format: "Model Name" by Creator Name, licensed under CC BY 4.0, via Sketchfab
Importing Models
Download Process
Importing into Blender
- FBX Import
- OBJ Import
- glTF/GLB Import
Best for Blender - includes materials and textures
File→Import→FBX (.fbx)- Navigate to downloaded file
- Adjust import settings:
- ✅ Enable "Import Materials"
- ✅ Enable "Import Normals"
- Consider scale (some models may import at incorrect size)
Universal format - good compatibility
File→Import→Wavefront (.obj)- Navigate to downloaded file
- Note: Materials and textures may need manual connection
- Check UV maps in UV Editing workspace
Modern format - good for web-based reconstructions
File→Import→glTF 2.0 (.glb/.gltf)- Navigate to downloaded file
- Materials usually import correctly
- Ideal if you plan to export back to web viewers
Post-import checklist:
- Check scale against your scene (use measuring tools)
- Verify materials and textures loaded correctly
- Assess polygon density - decimate or retopologize if too high
- Check normals (blue faces indicate inverted normals)
Practical Applications
Use Case: Contextual Scene Building
Sourcing models to populate a scene around your primary reconstruction:
Contextual Scene Building
Sourcing models to populate a scene around your primary reconstruction:
- Urban furniture (benches, trash bins, signs)
- Vegetation (trees, bushes appropriate to region)
- Vehicles (period and region appropriate)
- Architectural context (neighboring buildings)
Reference Matching
Using existing models as reference for your own modeling:
- Find similar objects or structures
- Use as scale and proportion reference
- Study topology for modeling techniques
- Compare with your source imagery
Workflow
- Complete your primary reconstruction first
- Identify gaps that need contextual elements
- Search Sketchfab with specific terms
- Download and import selected models
- Scale and position appropriately
- Document all sourced assets
Summary
Sourcing 3D models from platforms like Sketchfab is a powerful technique for accelerating visual investigation work. By understanding licensing requirements, evaluating model quality, and properly integrating sourced assets into your workflow, you can build comprehensive 3D reconstructions more efficiently without compromising accuracy or legal compliance.
Remember that sourced models should complement your investigation work, not replace it. Key elements specific to your investigation should still be modeled from your source materials, while contextual and supporting elements can be sourced to provide necessary environmental context.
Key Takeaways:
- Sketchfab provides access to millions of downloadable 3D models with clear licensing
- Verify and document the license type before using any sourced model
- Assess model quality by checking polygon count, texture resolution, and accuracy
- Properly attribute all sourced models in your documentation and final outputs
- Use sourced models strategically for context while modeling key elements yourself
Documentation for Methodology
To ensure your work aligns with the Berkeley Protocol and the Guide for Judges, use the following points to document your use of sourced models.
Methodology: How to document Sourced Models
In Your Method Section
Guidance: Describe the purpose, the tool version, and the input data.
- Investigative Objective: To provide contextual or reference geometry for the scene using third-party assets.
- Software Version: Sketchfab / [Source Platform].
- Input Data: Model [Name/ID] by [Creator] (Source URL/Hash).
In Your Decision Log
Guidance: Record the specific procedural parameters and integrity checks.
- Specific Settings: License ([CC-BY/CC0]), File Format ([FBX/OBJ]), Scale Factor ([X.X]).
- Assumptions: Assumed the sourced model was a representative approximation of the real-world object, not a forensic replica.
- Integrity Check: Verified that the model's license permitted use in the investigation and that attribution was properly recorded.
- Date Performed: [YYYY-MM-DD].
Verification & Mitigation
Guidance: How did you mitigate bias or verify accuracy?
- Verification: Compared the sourced model's dimensions with known specifications or reference imagery to ensure scale accuracy.
- Bias Mitigation: Clearly distinguished between "evidence" (modeled from data) and "context" (sourced assets) in the final scene to avoid misleading the viewer.
Common Limitations
- Sourced models may contain artistic inaccuracies, incorrect topology, or texture errors.
- Scale is often arbitrary and must be manually verified and corrected.
Example Methodology Statement
"Contextual vehicles were sourced from Sketchfab (User: [Name], License: CC-BY 4.0). The model [Model Name] was imported and scaled to match the dimensions of a [Real Vehicle Year/Make/Model] (Length: [X]m). These assets are for illustrative context only and were not used for spatial analysis."
Further Resources:
- Sketchfab - Main platform for 3D model discovery
- Creative Commons License Guide - Understanding CC licenses
- Sketchfab Help Center - Platform documentation
- OpenHeritage3D - Cultural heritage 3D models
- Blender Import Documentation - File format import guides