Depth Anything 3

Depth Anything 3

A demo Space for Depth Anything 3, a research model that, going by its title, recovers a scene's spatial structure from images shot from any viewpoint. The README holds only a citation.

Made by depth-anything on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: Depth Anything 3 is a demo Space for a research model that, going by its paper title, recovers a scene's spatial structure from images taken from any viewpoint. It runs the DA3NESTED-GIANT-LARGE model on ZeroGPU hardware and also names facebook/map-anything. The README holds only a citation, so the interface details come from opening the page.

Who it suits: Computer vision researchers, 3D developers and students who want to try a recent depth model on their own images will get the most from it. The rows below are judgment from the name and the files, not claims from the authors.

Industry | Who uses it | How they use it | Example
Research | Computer vision researchers | Try the model on their own images before reading the paper | Checking how it handles a photo from an odd angle
Robotics | Robotics engineers | Judge whether the depth output suits their scenes | Running a few camera frames through the demo
3D and graphics | 3D artists and developers | See how a scene's layout is recovered from several photos | Uploading a few photos of the same object
Education | Students | Watch a recent depth model work on real pictures | Comparing results on near and far objects

You can skip it if you want a documented walkthrough, since the README holds only a citation.

Where to use it: Use it in a browser tab on Hugging Face, or run a copy on your own computer.

• A web browser: open the Space page on Hugging Face, where it runs on ZeroGPU
• Windows, macOS or Linux: run a copy with Python 3.11

Use case | Who | What they get back
Try the model online | Researchers and developers | The model's result for your images, as the interface shows it
Study the code | Learners | The depth_anything_3 folder and app.py to read
Run a local copy | Developers with Python 3.11 | The same Gradio app on their own machine

How to set it up: Using the Space online needs nothing. A local copy needs Python 3.11, the listed packages and the system packages in packages.txt, which are installed with apt-get.

• Python 3.11 with pip
• Gradio 5.49.1
• The packages in requirements.txt
• The system packages in packages.txt

1. Install the Hugging Face hub package.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Install Gradio 5.49.1, then the listed packages.
5. Start the app with Python.

pip install -U huggingface_hub
hf download depth-anything/depth-anything-3 --repo-type space --local-dir depth-anything-3
cd depth-anything-3
pip install gradio==5.49.1
pip install -r requirements.txt
python app.py

The repository also holds requirements-basic.txt, but the source gives no command for it, and it states no hardware need for a local run.

How to use it: Open the Space page and use the interface you find there. The README gives no steps. With a local copy, start the app file and use the interface it opens.

python app.py

Example: Open the Space, give it a photo or a few photos of the same room, and run it. What comes back should be a depth result for the scene, though the README shows no sample output.

For newcomers and experts: Depth estimation means working out how far each part of a picture is from the camera. Start with one clear photo that has both near and far objects, and see what the interface returns.

The record names two models and one dataset, and the nested giant-large name suggests a large variant, but the README gives no sizes or scores. The model code ships in the repository as the depth_anything_3 folder, next to a pyproject.toml.

You give the app your images
The Space requests GPU time on ZeroGPU
The Depth Anything 3 model processes the views
The result appears in the interface

Good to know:

• License: cc-by-nc-4, a non-commercial Creative Commons license
• Cost to run: no API key or paid service named, and the Space uses zero-a10g hardware at Hugging Face
• Data it reaches: the images you give it, and the source names no other service
• Citing: the README asks research users to cite the Depth Anything 3 paper
• Not for: commercial use, since the license is non-commercial, and the README states no other limits

On Hugging Face, 13 November 2026

• Likes: 490
• Built with: gradio
• License: cc-by-nc-4.0

Links

• Space: https://huggingface.co/spaces/depth-anything/depth-anything-3
• Files: https://huggingface.co/spaces/depth-anything/depth-anything-3/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.