What Is CogVideo? Open-Source AI Video Model Explained
CogVideo and CogVideoX are open-source text-to-video and image-to-video AI models from THUDM and Zhipu AI. Learn what they are, how they differ, the model family, and how to use them online.
CogVideo is a family of open-source AI models for generating video from text and from images. It started as the original CogVideo research model in 2022 and grew into the modern CogVideoX family in 2024. This page explains what each one is, how they relate, and how to use them without writing any code.
Want to generate now?
You do not need to download the model or own a GPU. Open Text to Video or Image to Video and start in your browser.
What is CogVideo?
CogVideo is a large text-to-video generation model first released in 2022 by researchers at THUDM (Tsinghua University) and Zhipu AI. It was introduced in the paper CogVideo: Large-scale Pretraining for Text-to-Video Generation via Transformers and presented at ICLR 2023. At the time it was one of the largest and best-known open text-to-video transformers, and the name has since become an umbrella term for the project and the newer CogVideoX models.
The core idea is simple to state: you give the model a written description, and it produces a short video clip that matches it. Everything that followed builds on that foundation.
What is CogVideoX?
CogVideoX is the 2024 generation of the project and the version most people mean today. It is a diffusion-transformer model described in the paper CogVideoX: Text-to-Video Diffusion Models with an Expert Transformer. Compared with the original CogVideo, it generates longer, higher-resolution, more coherent clips and adds an image-to-video variant, so you can animate a still picture as well as generate from text alone.
CogVideoX is open source. The code is published on GitHub and the model weights are available on Hugging Face, which is why you see it run both in hosted tools like this one and on people's own machines.
How CogVideo and CogVideoX differ
It helps to keep the two names straight:
- CogVideo (2022) is the original research model. It established the approach and the brand.
- CogVideoX (2024) is the current family. It is what powers modern text-to-video and image-to-video generation, and it is what you are using when you generate a clip here.
In everyday use, people say "CogVideo" to mean the whole project and "CogVideoX" to mean the specific current models. When a guide talks about VRAM, model sizes, or the 2B and 5B variants, it is talking about CogVideoX.
The CogVideoX model family
CogVideoX ships in a few sizes and variants. The smaller model is lighter to run, and the larger model produces higher quality.
| Model | Size | Task | Notes |
|---|---|---|---|
| CogVideoX-2B | ~2B parameters | Text to Video | Lightest to run locally. Lowest VRAM. |
| CogVideoX-5B | ~5B parameters | Text to Video | Higher quality than 2B. |
| CogVideoX-5B-I2V | ~5B parameters | Image to Video | Animates a still image into a clip. |
| CogVideoX1.5-5B | ~5B parameters | Text and Image to Video | Later release targeting higher resolution and longer clips. |
If you plan to run a model yourself, the system requirements guide lists the exact VRAM and GPU each variant needs.
Is CogVideo open source?
Yes. The CogVideoX code is open source on GitHub, and the model weights are published openly on Hugging Face. That means anyone can read the code, download the weights, and run the models on their own hardware, which is unusual for a video model of this quality and a big part of why CogVideo is so widely searched and discussed.
Being open source also means you have two ways to use it. You can set it up locally if you have a capable NVIDIA GPU, or you can use a hosted version like CogVideo Online and skip the setup entirely.
Use CogVideo without installing anything
Running a video model locally needs a strong GPU and a working Python environment. If you would rather just make a video, the hosted tools run CogVideoX for you in the browser.
Getting Started
Make your first video in a few minutes.
Text to Video
Generate a clip from a written description.
Image to Video
Animate a still image into motion.
System Requirements
VRAM and GPU needed to run CogVideoX locally.
Frequently asked questions
Is CogVideo free?
Browsing the tools is free, and you can use CogVideo Online to generate video in your browser. Starting a generation requires signing in and uses credits. To run the open-source model on your own hardware, the weights are free to download.
What is the difference between CogVideo and CogVideoX?
CogVideo is the original 2022 model and the name of the wider project. CogVideoX is the 2024 family of models that powers current text-to-video and image-to-video generation. When you generate a clip today, you are using CogVideoX.
Who created CogVideo?
CogVideo and CogVideoX come from THUDM (Tsinghua University) and Zhipu AI. The work is published in peer-reviewed papers and released as open source.
Can I run CogVideoX on my own computer?
Yes, if you have a capable NVIDIA GPU. CogVideoX-2B is the lightest to run, and the larger 5B models need more VRAM. See the system requirements guide for the exact numbers, or use CogVideo Online to skip local setup.
Does CogVideo support image to video?
Yes. CogVideoX-5B-I2V animates a still image into a moving clip. You can try it on the Image to Video tool.