OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

GENIE2 Pytorch

GENIE2 Pytorch is a Python library implementing a Genie 2-style world-model framework in PyTorch. Its documented examples train through a loss on video and action inputs, generate future frames from an initial frame, and support interactive generation steered by action IDs.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

GENIE2 Pytorch — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when you want a code-level starting point for experiments with action-conditioned video generation and can work with PyTorch tensors and an implementation identified as work in progress.

Who it is for

  • Researchers experimenting with video-and-action world-model training.
  • Developers prototyping action-conditioned future-frame generation or interactive world-model demos.

Documented uses

  • Train a model using video and corresponding action IDs.
  • Generate future frames from an initial video frame.
  • Explore an interactive sequence in which action IDs guide successive frames.
Before you choose

Requirements & limits

The repository is public and lists an MIT license. Its README provides installation via pip.

  • This is a developer library, not a hosted generation app; the documented workflow is code-based.
  • The README's example uses synthetic tensors and does not describe preparing a real video-and-action dataset.
  • Interactive generation prompts for action IDs at successive frames.
  • The repository labels the implementation work in progress.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install the library and follow the documented generation example

Set up the Python package and inspect the documented training and frame-generation calls.

Before you start

  • Python environment with PyTorch available.
  • The README's sample inputs are PyTorch tensors; they are synthetic examples, not a real dataset.

Steps

  1. Install the package with the command below.

    Documented command
    pip install genie2-pytorch

  2. Import torch and Genie2, then create a model using the documented example parameters.

  3. Create video and action tensors, call the model to obtain a loss, and call loss.backward().

  4. Pass the initial frame to genie.generate with a frame count to request generated video.

Expected outcome

The README demonstrates a training-loss call and a generated video tensor containing the initial frame and requested future frames.

Guide limitations

  • The repository describes the implementation as work in progress.
  • The example uses synthetic tensors and does not document real-data preparation or evaluation.
Documentation 1 ↗
Source-based practical guide

Use the interactive generation example

Follow the documented pattern for prompting for action IDs as generated frames advance.

Before you start

  • Python environment with PyTorch and genie2-pytorch installed.
  • A Genie2 instance and video tensor shaped as in the README example.

Steps

  1. Create the model and example video and action tensors as shown in the README.

  2. Run the model call and backpropagation shown in the example.

  3. Call genie.generate with interactive=True and an init_action value.

  4. Enter the next action ID or IDs when prompted for each subsequent time frame.

Expected outcome

The example returns generated video and action values while prompting for actions during sequence generation.

Guide limitations

  • The interactive instructions are an example, not a documented integration with a game or simulator.
  • The implementation is labeled work in progress.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
16 October 2026
Original website title
lucidrains/genie2-pytorch · Official homepage title · retained in its source language
How we review sourcesSuggest a correction