H3 Camera Angle Control

H3 Camera Angle Control

Re-shoots a video clip or a photo from a new camera angle that you set by dragging a 3D camera, using the Meridian video model in a Gradio app.

Made by linoyts 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: H3 Camera Angle Control makes a new shot of a video or a photo from a viewpoint that was never filmed. You drag a 3D camera to choose the angle. Meridian, which is built by fine-tuning MiniMax-H3, works out the scene's 3D shape and then draws the shot again from your chosen viewpoint. It runs on ZeroGPU hardware at Hugging Face.

Who it suits: Video makers and photographers who want a different angle on a clip or still they already have.

Industry | Who uses it | How they use it | Example
Video production | Editors | Get an alternate angle for a shot | Orbit the camera around a subject in a clip
Product marketing | Content teams | Show a product from a new side | Move the camera around a studio photo
Photography | Photographers | Explore other viewpoints of one picture | Raise the camera above a still
Research | Vision researchers | Study view generation from 3D reconstruction | Read engine.py and camera3d.py

People who need exact, validated results for every angle can skip it, because only one example recipe is quality-validated.

Where to use it:

• A browser, by visiting the Space on Hugging Face
• A computer of your own that runs Windows, macOS or Linux and has Python 3.12.12
• A Hugging Face Space with ZeroGPU hardware

Use case | Who | What they get back
New angle on a video | Editors | A re-shot clip of 73 frames at 24 fps
New angle on a photo | Photographers | The final frame, a filmstrip and a downloadable MP4
Preview the warp | Anyone | A quick geometric projection with coverage shown

How to set it up:

• Python 3.12.12 and pip on Windows, macOS or Linux
• The Python packages in requirements.txt
• The system packages in packages.txt, which Hugging Face installs with apt-get
• Gradio 6.27.0, the version the Space uses
• Access to the gated facebook/VGGT-Omega checkpoint, or the buttons stay greyed out
• A GPU, since the Space is set to zero-a10g

1. Install the Hugging Face hub package.
2. Download the Space with the hf command.
3. Move into the new folder.
4. Install gradio at the version the Space uses, then the listed packages.
5. Start the app with Python.

pip install -U huggingface_hub
hf download linoyts/meridian-new-angles --repo-type space --local-dir meridian-new-angles
cd meridian-new-angles
pip install gradio==6.27.0
pip install -r requirements.txt
python app.py

The repository also holds a SETUP.md file with the full setup checklist.

How to use it: Online, visit the Space on Hugging Face, choose the Video or Image tab and add your source. Drag the camera widget, check the warp preview, then press Generate. On your own computer, run the start command and open the local page Gradio serves.

python app.py

Tool | What it does
Azimuth arc | Orbits the camera left or right, up to 60 degrees each way
Elevation arc | Drops the camera below or lifts it above, up to 30 degrees each way
Dolly line | Pushes in or pulls back, from 0.70x to 1.30x
Preview warp | Reruns the warp by hand after you type values
Generate | Makes the new-angle shot

Example: You load the sample bronze pharaoh bust and set a 30 degree orbit with the camera widget. After you press Generate, the app returns 73 frames of the bust seen from the new side, with a filmstrip and an MP4.

For newcomers and experts: If you are new, pick one of the three shipped examples. Choosing one fills in the whole recipe and runs nothing, so you can look at the controls before spending any GPU time.

If you know the field, three models split the work. VGGT-Omega makes a 3D version of the scene, which lets the source pixels be projected again through the new camera. Meridian's fine-tuned MiniMax-H3 teacher and its four-step DMD adapter generate the shot, and the base model's VAE encodes and decodes frames. Grey areas in the warp mean the source has no reference there.

You add a clip or a photo
VGGT-Omega rebuilds the scene in 3D
The warp shows what the source covers
Meridian generates the new shot
You get 73 frames and an MP4

Good to know:

• Input: a single image, or an unbroken video that runs at least 3.05 seconds past your chosen start, with a file size below 100 MB
• Artifacts: large camera moves show geometry the source never saw, so flaws are expected at the extremes
• Cost: each camera drag spends one geometry call on ZeroGPU xlarge, so the form locks while a warp runs
• License: apache-2 for the app code, while model and third-party media licenses are separate
• Cost to run: GPU hardware and access to the gated VGGT-Omega checkpoint, no paid service is named
• Data it reaches: the footage you upload, kept only temporarily, plus the models it fetches from Hugging Face
• Not for: sensitive footage, since the author warns against uploading it

On Hugging Face, 13 November 2026

• Likes: 18
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/linoyts/meridian-new-angles
• Files: https://huggingface.co/spaces/linoyts/meridian-new-angles/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.