4DAnyone Multi-View Video

4DAnyone Multi-View Video

Turns one portrait video into synchronized videos of the same performance from several camera views. A demo of the 4DAnyone model for researchers building 4D reconstruction inputs.

Made by hugging-apps 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: 4DAnyone Multi-View Video is an interactive demo of the 4DAnyone model. You give it one portrait video from a single camera, and it returns several videos of the same performance as seen from different camera positions.

The README says these views are the kind of input a 4D Gaussian-splatting reconstruction needs. The output is H.264 videos, a camera rig file named cameras.json and run metadata, all zipped for download.

Who it suits: Researchers and developers who work on 4D reconstruction, avatars or multi-view video and want to run the model without building the stack.

Industry | Who uses it | How they use it | Example
Research | 4D reconstruction researchers | Make multi-view input from one video | Run a portrait clip and download the zip
Visual effects | Technical artists | Test camera-ring views of a performance | Generate four views of one take
Software | Computer vision developers | Evaluate the model before integrating it | Compare runs with fewer and more views

You can skip it if you have no portrait video of one person, or if you need commercial use, since parts of the pipeline are research-only.

Where to use it:

• A web browser, on the Space's page at Hugging Face, with nothing to install
• Windows, macOS or Linux with Python, from a copy you download, on a GPU with a lot of memory
• An MCP client, since the Space carries an MCP tag, though the source gives no setup for it

Use case | Who | What they get back
Multi-view from one clip | Researchers | Up to six synchronized videos of the performance
Camera rig for reconstruction | 3D developers | A cameras.json file in OpenCV style
Quick check of the model | Engineers | A zip with videos and run metadata

How to set it up:

• Python 3.12 on your computer
• The Python packages listed in requirements.txt
• The system packages listed in packages.txt, which the Space installs with apt-get
• A GPU with large memory, since a four-view run peaks at 53 GiB of VRAM
• The huggingface_hub package, which gives you the hf command

1. Install the huggingface_hub package.
2. Download the Space with the hf command.
3. Move into the downloaded folder.
4. Install the pinned Gradio version.
5. Install the packages from requirements.txt.
6. Start the app with Python.

pip install -U huggingface_hub
hf download hugging-apps/4danyone-multiview-demo --repo-type space --local-dir 4danyone-multiview-demo
cd 4danyone-multiview-demo
pip install gradio==6.17.3
pip install -r requirements.txt
python app.py

How to use it: Open the Space's page in your browser, or start your own copy with the command below. Add a portrait video of one person, choose the start time and the number of views, run it and download the zip.

python app.py

Example: You add a vertical clip of one person talking and ask for four views. After about nine minutes of ZeroGPU time, you get a zip with four H.264 videos of the same performance plus cameras.json.

For newcomers and experts: If you are new to multi-view video, start with a short vertical clip of one person and ask for fewer views, because fewer views run faster. Look at how the views line up.

If you know the field, the pipeline decodes 121 frames, recovers motion with GVHMR, mattes the subject with BiRefNet and renders a skeleton from a virtual camera ring. A video DiT derived from Wan2.2-TI2V-5B then denoises all views together. To fit ZeroGPU, a precomputed embedding of one fixed prompt replaces the 11.4 GB text encoder.

Decode a 121-frame clip
Recover body motion with GVHMR
Matte the subject and render a skeleton
Generate the views with a video DiT
Export H.264 videos and a camera rig

Good to know:

• License: the Space lists apache-2, and the README says the model and code are Apache-2.0
• Cost to run: a PRO-tier or higher ZeroGPU quota at Hugging Face, since a four-view run takes about nine minutes
• Data it reaches: your input video, plus GVHMR code cloned when it runs and the model weights
• Not for: commercial use of every part, since the README marks GVHMR as research use only, SMPL-X as research licensed and YOLOv8x as AGPL-3.0
• Input limits: a vertical clip of one person, close to 9:16, with 121 frames left after your chosen start, and no more than six views

On Hugging Face, 13 November 2026

• Likes: 47
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/hugging-apps/4danyone-multiview-demo
• Files: https://huggingface.co/spaces/hugging-apps/4danyone-multiview-demo/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.