Skip to content

Pippo

Generate 1K multi-view video from a single photo

Visit official site

What Pippo is

The model is described as a generative model developed by Meta Reality Labs with several universities, producing high-resolution multi-view videos from a single ordinary photo. It needs no extra input such as parametric models or camera parameters to generate 1K resolution video, and it is based on a multi-view diffusion transformer architecture.

What you can do with it

  • Generate high-resolution multi-view video from a single photo
  • Train models at 128, 512 and 1024 resolutions
  • Use sample training code and dataset support such as Ava-256
  • Compute reprojection error between generated and real images
  • Apply control MLP and attention bias techniques to the diffusion transformer
  • Run under different GPU configurations such as A100 and T4

Application scenarios

  • Researchers generate multi-view video from a single photo for virtual reality content
  • Film teams generate high-resolution virtual character video
  • Developers extend the code architecture into new image generation applications

Getting started

  • Clone the repository from its GitHub address
  • Create a Conda environment and install dependencies such as PyTorch
  • Run the sample download script to fetch Ava-256 data
  • Pick a model config for the GPU setup and start training

What to watch out for

  • The code is open source but pre-trained weights are not included
  • Users need to train the model themselves

Pros & cons

✓ What we like

  • No parametric model or camera parameters are needed
  • Resolutions 128, 512 and 1024 are available for training

! What to watch out for

  • Pre-trained weights are not included with the open-source code
  • Users must train the model on their own

FAQ

What does Pippo generate?

The material describes high-resolution multi-view video generated from a single ordinary photo.

Are pre-trained weights provided?

The material states the code is open source but pre-trained weights are not included, so users train the model themselves.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review