thu-ml

thu-ml / Causal-Forcing

Public

Official codebase for "Causal Forcing: Autoregressive Diffusion Distillation Done Right for High-Quality Real-Time Interactive Video Generation"

thu-ml.github.ioCausalForcing.github.io auto-regressive-diffusion-model autoregressive-models consistency-models diffusion diffusion-models
403
18
100% credibility
Found Feb 03, 2026 at 163 stars 2x -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Causal Forcing is a research codebase for training and deploying autoregressive diffusion models that generate high-quality real-time videos from text prompts.

How It Works

1
🔍 Discover video magic

You stumble upon a cool project that turns words into smooth, lively videos right on your computer.

2
📥 Grab the ready pieces

Download the smart helpers (like pre-made brains) that make video creation easy and fast.

3
🚀 Start your video maker

Run a simple command to wake up the tool and connect it to your ideas.

4
✨ Describe your dream scene

Type a fun description like 'a cat dancing on the beach at sunset' and hit go – watch frames appear live!

5
🎬 Make longer stories

Extend to minute-long clips by sliding in more scenes seamlessly.

🎉 Share your masterpiece

Your high-quality, real-time video is ready to wow friends – smooth motion, perfect details, all from text!

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 163 to 403 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is Causal-Forcing?

Causal-Forcing is the official GitHub repository for a Python codebase implementing causal diffusion forcing, an autoregressive diffusion distillation technique for high-quality real-time interactive video generation. It generates streaming videos from text prompts with sharp visuals and smooth motion dynamics, outperforming baselines like Self Forcing while matching their inference speed on a single RTX 4090. Users run CLI inference via scripts like inference.py for frame-wise or chunk-wise modes, train custom models, or launch a web demo for instant previews.

Why is it gaining traction?

It delivers done-right autoregressive causal forcing, fixing quality drops in video distillation without hiking compute costs—key for real-time apps. Pretrained models on Hugging Face, official GitHub releases, and long-video extensions via Rolling Forcing hook devs needing efficient text-to-video pipelines. The chunk-wise stability and frame-wise expressiveness give tangible wins over multi-step diffusion baselines.

Who should use this?

AI researchers tuning diffusion models for causal video synthesis, video engineers building interactive streaming tools like live avatars or AR previews, or indie devs prototyping text-to-video in Python apps.

Verdict

Worth forking for video gen experiments—download official GitHub release models and test inference.py today. At 296 stars and 1.0% credibility score, it's nascent with solid docs but light on tests; polish for prod use.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.