thu-ml/DiT-Extrapolation
Official implementation for "RIFLEx: A Free Lunch for Length Extrapolation in Video Diffusion Transformers" (ICML 2025) , UltraViCo (ICLR 2026) and UltraImage
This project helps video creators and digital artists extend the duration of existing AI-generated videos, turning short clips into longer, more detailed sequences. It takes an existing short video, often generated by AI models like HunyuanVideo or CogVideoX, and produces a significantly longer version while maintaining visual quality and consistency. Digital content creators, animators, and AI artists who use diffusion models for video generation are the primary users.
794 stars.
Use this if you need to create longer, high-quality videos from shorter AI-generated clips or generate high-resolution images, specifically when working with diffusion transformer models.
Not ideal if you need to create videos from scratch without leveraging existing AI models or require real-time video generation for live applications.
Stars
794
Forks
74
Language
Python
License
Apache-2.0
Category
Last pushed
Mar 08, 2026
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/thu-ml/DiT-Extrapolation"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Related models
hao-ai-lab/FastVideo
A unified inference and post-training framework for accelerated video generation.
ModelTC/LightX2V
Light Image Video Generation Inference Framework
thu-ml/TurboDiffusion
TurboDiffusion: 100–200× Acceleration for Video Diffusion Models
PKU-YuanGroup/Helios
Helios: Real Real-Time Long Video Generation Model
PKU-YuanGroup/MagicTime
[TPAMI 2025🔥] MagicTime: Time-lapse Video Generation Models as Metamorphic Simulators