# tianweiy/CausVid

(CVPR 2025) From Slow Bidirectional to Fast Autoregressive Video Diffusion Models

Repository: https://github.com/tianweiy/CausVid
Canonical: https://ross.abutalabs.com/products/causvid
Language: Python
License: NOASSERTION
License Family: other
Last push: 2025-08-07T04:49:23+00:00

## Health v2 (maintenance only)
Score: 36/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 35, release rhythm 35, longevity 38
- inputs: {"age_days": 537, "days_push": 391, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1426, forks 88 (observed 2026-08-28T04:04:41.669535+00:00)

## What it is
CausVid is a research codebase implementing a fast autoregressive video diffusion model distilled from a bidirectional diffusion transformer, enabling few-step text-to-video generation. It supports streaming generation at ~9.4 FPS on a single GPU, plus zero-shot video-to-video translation, image-to-video, and dynamic prompting.

## Use cases
- generate videos from text prompts in a few steps
- stream high-quality video generation in real time on a single GPU
- translate video to video with a diffusion model
- generate video from a single image
- change prompts dynamically during streaming video generation
- distill a 50-step video diffusion model into a 4-step generator

## When to choose
- you need fast, few-step text-to-video generation rather than slow multi-step diffusion
- you want streaming/autoregressive video synthesis with low latency
- you are researching video diffusion distillation or causal attention in diffusion transformers

## When to avoid
- you need a polished production-ready video generation product rather than research code
- you lack a CUDA GPU or cannot download the Wan2.1 base model checkpoints
- you need a stable API since the repo is a work in progress with frequent updates

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, video-processing, llm-inference
- domain: deep-learning, artificial-intelligence, computer-vision
- platform: python
- tags: video-diffusion, text-to-video, distillation, autoregressive, diffusion-transformer, streaming-generation, research-code, video, gpu, linux

## Member repositories
- tianweiy/CausVid (main) score 36

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:41.669535+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:37:29.001674+00:00, confidence not recorded.
  - readme: https://github.com/tianweiy/CausVid (fetched 2026-08-28T04:04:41.669535+00:00, sha 1cbe93f2073a)
- Data as of 2026-08-30T08:39:29.467469+00:00.