lucidrains/make-a-video-pytorch

Health v2 · maintenance only

Adoption not part of the score

1986 stars · 185 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A PyTorch library implementing Make-A-Video, Meta AI's text-to-video generation approach, built around pseudo-3d (axial) convolutions and spatiotemporal attention modules. It provides reusable building blocks (PseudoConv3d, SpatioTemporalAttention) that work on both video and image tensors so models can be pretrained on images and ported to video.

Use cases

• generate videos from text prompts
• implement text-to-video diffusion models in pytorch
• add temporal attention to a 2d unet
• pretrain on images then fine-tune for video generation
• use pseudo-3d convolutions for spatiotemporal modeling
• research frame interpolation for video synthesis

When to choose

• you need the Make-A-Video architecture components in PyTorch
• you want to extend a text-to-image model to video generation
• you are doing research on spatiotemporal attention and axial convolutions

When to avoid

• you need a ready-to-use, fully trained text-to-video model
• you want a production video generation pipeline with inference tooling
• you need a maintained, stable implementation - the repo is marked work-in-progress

Facets

library · maturity experimental

deep-learning machine-learning image-processing video-processing deep-learning machine-learning artificial-intelligence python text-to-video diffusion pytorch pseudo-3d-convolutions spatiotemporal-attention research-implementation generative-ai video

Member repositories

For agents

markdown · JSON · MCP: product_card(name="lucidrains/make-a-video-pytorch")