PRISM Image to 3D

PRISM Image to 3D

Builds a 3D Gaussian scene from a single picture and shows it in a 3D viewer. It suits researchers and 3D hobbyists who want to try single-image reconstruction without installing anything.

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: This Space is a Gradio demo of PRISM, a method that rebuilds a 3D scene from one image in a single forward pass. The result is a set of 3D Gaussians, shown in a viewer. It runs on zero-a10g hardware at Hugging Face and is tagged MCP.

Who it suits: It suits people who want to see single-image 3D reconstruction work, and researchers who want to read how the pieces are chained together.

Industry | Who uses it | How they use it | Example
Computer vision research | Researchers | Test single-image 3D reconstruction | Run an example image and inspect the Gaussian scene
3D and graphics | 3D hobbyists | Turn a picture into a viewable scene | Look at a scene from several angles in the viewer
Software | Developers | Study a Gradio demo of a research model | Read app.py and the prism folder

You can skip it if you need a finished, production-quality 3D model, since this is a research demo.

Where to use it: You can use it in these places:

• A browser: open the Space's page on Hugging Face and nothing needs installing.
• Windows, macOS or Linux: run your own copy with Python 3.12.

Use case | Who | What they get back
Single-image 3D | Researchers | A 3D Gaussian scene predicted from one picture
Scene viewing | Hobbyists | A viewer showing a subset of the predicted Gaussians
Pipeline study | Developers | A working chain from depth to rendered Gaussians

How to set it up: You need these first for a local copy:

• Python 3.12 and pip
• A GPU, torch and CUDA setup that fits the gsplat wheel kept in the wheels folder

The online page needs none of this. For your own copy, follow the steps below.

1. Install the Hugging Face tools.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Install Gradio at the version the Space uses, then the other packages.
5. Start the app.

pip install -U huggingface_hub
hf download hugging-apps/prism-image-to-3d --repo-type space --local-dir prism-image-to-3d
cd prism-image-to-3d
pip install gradio==6.31.0
pip install -r requirements.txt
python app.py

How to use it: Open the Space in a browser, or start your own copy with the command below. Then choose an image and run the reconstruction.

python app.py

Tool | What it does
MoGe | Estimates depth from the input image
Cosmos tokenizer | Encodes the input into a latent
PRISM residual encoder | Works with a latent prior warped along six camera paths
Lyra 3DGS decoder | Decodes the latent into 3D Gaussians
gsplat | Renders the Gaussians for the viewer

Example: You pick one of the example images, or your own picture, and start the run. The Space predicts a 3D Gaussian scene and shows part of it in the viewer, with the most opaque Gaussians first.

For newcomers and experts: If you are new to this, open the Space and try an example image first. Turn the scene in the viewer to see depth that a flat picture does not show.

If you know the field, the chain runs depth from MoGe, then a Cosmos tokenizer latent, then a forward-warped latent prior with a PRISM residual encoder. A frozen Lyra decoder turns that into Gaussians, which gsplat renders. The repository bundles a gsplat wheel because ZeroGPU has no prebuilt one, and vendors the Triton kernels from mamba-ssm.

Image goes in
MoGe estimates depth
Cosmos latent is warped six ways
Lyra decoder makes Gaussians
gsplat renders the viewer scene

Good to know: The viewer shows only a subset of the predicted Gaussians, the most opaque first. The Lyra decoder weights are downloaded at runtime from the nvidia/Lyra repository.

• License: apache-2 for the PRISM and Lyra code, NVIDIA Open Model License for the Lyra decoder weights
• Cost to run: none named
• Data it reaches: model weights downloaded from Hugging Face at runtime, and the example images from the nvidia/Lyra-Testing-Example dataset under CC-BY-4.0

On Hugging Face, 13 November 2026

• Likes: 6
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/hugging-apps/prism-image-to-3d
• Files: https://huggingface.co/spaces/hugging-apps/prism-image-to-3d/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.