Sam3D
Single image 3D reconstruction modelTurn one photo into a 3D asset
Highlights
What you can do with it
- Creative production scanning a prop from one photo for a game engine
- E-commerce AR showing a product in a room from a single image
- Robot perception inferring shape and free space from camera images
How to use it
- 1 Capture a well lit RGB image of the target object
- 2 Optionally use SAM 3 text or box prompts to isolate it
- 3 Run inference with the published checkpoints and code
- 4 Export the mesh and texture into your target pipeline
- 5 Use a clean background and balanced light for better masks
Who it is for
Pros
- One photo instead of a multi-view or LiDAR setup
- Open weights, code and benchmarks aid reproducibility
- Prompt based segmentation makes targeting flexible
- Free and open source
Limitations
- Research level, so production quality is not guaranteed
- Output quality depends heavily on lighting and occlusion
- Running inference needs suitable hardware and setup
- Free open source release does not imply ongoing support
What Sam3D is
Sam3D is a research level 3D reconstruction model from Meta that works from a single RGB image. It pairs the open vocabulary segmentation of SAM 3 with geometry, texture and layout prediction, so a photo goes in and a 3D asset comes out. The listing notes that checkpoints, inference code and benchmark datasets such as Artist Objects and SAM 3D Body are released, which matters for anyone who needs to reproduce a result. It is described as free and open source, with applications in creative production, e-commerce AR, robot perception and scientific visualisation.
What you can do with it
- Turn one product or prop photo into a mesh with textures.
- Use text, point or box prompts to isolate the object you care about.
- Feed the result into a 3D engine or an AR viewer.
- Give a robot a shape and free space prior when depth data is missing.
- Convert 2D scans or microscope images into 3D form for inspection.
Who it is for
- Creative teams producing games, CGI and social content.
- E-commerce builders adding view in room AR features.
- Robot researchers who need 3D priors from camera input.
- Medical and scientific researchers working from 2D imagery.
What to watch out for
This is a research level release, so output quality will not always match a hand built asset and some meshes will need cleanup in Blender or a game engine. Results depend strongly on the input: clear images, balanced lighting and minimal occlusion give better masks and geometry, while busy backgrounds hurt accuracy. Running inference needs suitable hardware and some setup, and the listing does not promise turnkey performance. Being free and open source means no guaranteed support, so budget time for experimentation, and validate any asset before it goes into a product or a published study.
FAQ
What does Sam3D do?
It reconstructs 3D shape, texture and layout from a single RGB image.
Who released it?
The listing describes it as a research level model from Meta.
Is it open source?
Yes, the listing says checkpoints, inference code and benchmark datasets are released and the model is free and open source.
How do I isolate an object?
You can use SAM 3 prompts with text, points or boxes.
What improves the result?
Clear images, balanced lighting, minimal occlusion and a simple background help mask quality and geometric accuracy.
Why it is free
What the official page says： 免费
Free limits, trial length and features are decided by the official site and can change at any time.
Product preview
Screenshot archived from the listing page, for preview only.
