Photogram3D, a Blender 5.2 add-on that generates meshes from photos and video, is released

What happened

Photogram3D is a Blender 5.2 add-on that creates textured meshes from photos or video. It is distributed on Gumroad (priced at $0+, meaning from free to any amount the user chooses), and the source is published in a GitHub repository. The version listed on the page is photogram3d-0.3.15.

How it works

The workflow runs in this order: photos or video, camera pose estimation, dense point cloud, cropping inside Blender, mesh generation, and texture generation with blended seams. COLMAP 4.2.1 is used for camera estimation, and OpenMVS 2.4.0 is used to generate depth maps, meshes and textures. Operation is done from the 3D Viewport sidebar (N → Photogram3D). In the crop stage, users can level the ground, set real-world scale by specifying two points, and place a crop box.\n\nQuality presets range from Draft to Full, and one is suggested based on the detected CPU, RAM and GPU. Depth maps can be computed on the CPU, on an NVIDIA GPU, or on both at the same time. For video input, frames are selected based on camera motion, a camera that follows the footage is created for each frame, and the video can be played behind the scan, according to the article.\n\nFeatures for difficult capture sets include a path check for folded camera solutions, solution estimation that follows the shooting order (for objects that look similar from the opposite side), AI Features (ALIKED and LightGlue) for objects with little texture, re-solving without selected cameras, and re-solving from existing cameras. There is also a function to merge two scans (such as two videos, or captures made on different days) into one. Additional features include AI Polish, which uses MoGe-2 normals

By the numbers

Why it matters

With photogrammetry, users have had to operate COLMAP and OpenMVS separately and then bring the results into a DCC tool. This add-on consolidates those steps in the Blender sidebar and is said to automate tool download and setup as well. For creators who use Blender, and for people who want to try scanning artifacts or small objects, it may lower the barrier to getting started.\n\nOn the other hand, support is limited to Blender 5.2 and mainly Windows environments. For business use in Japan, users need to check whether it runs on their own hardware and OS, and whether the GPL-3.0-or-later terms suit their intended use.

How to try it

1. In Blender 5.2, go to Edit › Preferences › Get Extensions, choose "Install from Disk…", and select the zip downloaded from Gumroad.
2. Enable Allow Online Access under Preferences › System › Network.
3. In the 3D Viewport, open N → Photogram3D and run Install Tools (about 0.6 GB download).
4. Start by trying a photo set of a small subject with a lower-quality preset.

Caveats and open questions

• Benchmarks showing accuracy, speed or reproduction quality cannot be confirmed from the material provided.
• Testing is stated to have been done on only two Windows laptops. Operation on Linux and macOS, including installation of external tools, has not been confirmed.
• The author credit includes "Anthropic Claude AI". The development background and measurement methods are said to be described in DEVELOPMENT.md, but the material does not include its contents.
• The body of the GitHub repository could not be retrieved, and the README, the full license text and the change log have not been checked.
• The quality and output formats of the Gaussian splat feature using Brush cannot be confirmed from the material.
• The license terms of the bundled tools, such as COLMAP and OpenMVS, are not stated in the material.

Glossary

Photogrammetry

A method of reconstructing the 3D shape of an object from multiple photos.

COLMAP

Open-source software that estimates camera positions and orientations from photos and builds point clouds.

OpenMVS

A library that reconstructs dense geometry from estimated cameras and creates meshes and textures.

Depth map

An image that represents the depth to each pixel.

ALIKED + LightGlue

A combination of AI methods for detecting and matching image feature points, said to be effective for objects with few patterns.

MoGe-2

A model that estimates geometric information from a single image. In this add-on, it is used to refine normals.

Brush

A tool for training and generating Gaussian splats.

Sources

Primary sources (publisher, paper, repository)

• Photogram3DGumroad (tricky_3d)（tricky3d.gumroad.com）・en
• iuriikruglov/photogram3dGitHub（github.com）・en

This article was generated in Japanese by AI from the sources listed (primarily primary sources) and published automatically, without prior human review, after automated checks such as source matching (how we use AI). Figures and specifications are as stated by the publishers and have not been independently verified by us. Please consult the original sources. To report an error, use our contact form; we will review it promptly and correct it. This page is an AI translation of the Japanese original. If anything differs, the Japanese version prevails. Japanese original