Cog
The open source AI video model
What is CogVideoX?
CogVideoX is the open source text to video model from THUDM. Run it here as CogVideo, online and free, with no GPU and nothing to install.
Free to start. No GPU required. Built on the open source CogVideoX from THUDM.
About CogVideoX
What CogVideoX Is, in Plain Terms
An open text to video model you can read, run, and build on, offered here with no local setup.
Open source, from THUDM
CogVideoX is released openly by THUDM, the research group at Tsinghua University. The weights and code are public, so it is inspectable technology rather than a black box.
No GPU needed here
Running CogVideoX locally needs an NVIDIA GPU with about 5 GB of VRAM or more. On cogvideo.net it runs on our servers, so any laptop or phone with a browser works.
Text and image to video
At its core CogVideoX generates video from a text prompt. An image to video variant animates a single still while keeping your subject and framing.
Free to start
Generate your first CogVideoX videos for free, with no credit card and no subscription just to try it.
No install, no setup
Skip Python, CUDA, and diffusers. Open the page, type a prompt or upload an image, and generate straight in your browser.
Ready in minutes
A clip renders on our servers in a minute or two, then downloads as a standard MP4 you can post or edit.
The model family
The CogVideoX Model Family
CogVideoX is a line of open models from THUDM, not a single model. Here is what each release does. On cogvideo.net the line powers the hosted CogVideo generator, so you use it with no install and no GPU.
CogVideoX-5B
Highest qualityThe larger open model in the family, tuned for the most detail and the most coherent motion. Reach for it when the final look matters more than raw speed.
CogVideoX-2B
LightestThe smaller, most efficient open model, built to run on modest hardware. The go-to when you want faster, lighter generations or plan to self-host on a smaller GPU.
See requirementsCogVideoX-5B-I2V
Image to videoThe image to video variant that takes a single still and animates it, keeping your subject and composition while adding natural motion.
Try image to videoCogVideoX 1.5
NewestThe newer release in the family, built for higher resolution and longer clips than the earlier CogVideoX models.
CogStudio
Community appA community app for running CogVideoX yourself, popular for image to video workflows. On cogvideo.net you get the same line fully hosted, with no setup.
CogVideo
The originalThe original open research model from THUDM that the whole CogVideoX line grew out of. It is the source project behind everything above.
Curious about VRAM, GPUs, and system requirements for running these models yourself? See the full VRAM and GPU guide
Three ways to create
Use CogVideoX Three Ways
Start from text, an image, or an existing clip. Each opens its own tool, free in the browser.
Text to Video
Describe a scene in words and CogVideoX generates a clip straight from the prompt.
Open Text to VideoImage to Video
Upload a still image and bring it to life with natural, believable motion.
Open Image to VideoVideo to Video
Restyle or transform an existing clip into something new with a written prompt.
Open Video to VideoHow it works
How to Use CogVideoX Online in Three Steps
Pick a CogVideoX tool
Choose the mode that fits your input: text to video for a scene from scratch, or image to video to animate a still. Both run the CogVideoX line as CogVideo, hosted for you.
Prompt or upload, then generate
Type what you want to see, or drop in your image, and press generate. The model runs on our servers, so there is no GPU to configure and no weights to download.
Review and download the MP4
Watch the result, then save a standard MP4 or run another take with a sharper prompt. Nothing about the CogVideoX setup is left on your machine.
Made with CogVideo
See What CogVideoX Creates
Every clip below was generated with CogVideo, from a short prompt or a single image.
What it is good at
What CogVideoX Is Well Suited For
Research to product
Because CogVideoX is an open, inspectable model, teams use it to move from a paper or a prototype to a shippable feature without licensing a closed engine.
Prototyping motion
Turn a written idea into a moving reference in minutes, so you can test how a shot reads before committing crew, budget, or a full CogVideoX self-host.
Animating a single image
The image to video variant is built to take one still and add believable motion, which suits product shots, portraits, and hero images.
Short social clips
Its short-clip output maps cleanly onto vertical formats, so a prompt becomes a Reels, Shorts, or TikTok-ready cut on a fast cadence.
Explaining an idea
Generate a quick illustrative clip for a concept, a lesson, or a pitch when a static slide or stock footage falls flat.
Campaign variations
Spin up several takes on one prompt to test hooks and directions, then carry the version that lands into a wider push.
FAQ
CogVideoX FAQ: Common Questions
Yes. CogVideoX is the open source text to video model line developed by THUDM, the research group at Tsinghua University. The weights and code are public, and cogvideo.net runs that same technology as a hosted service so you can use it without setting up anything locally.
You can start generating for free, with no credit card. Beyond the free credits, generations are priced per model, starting from 10 credits for CogVideo v1 and 100 credits for Veo 3.1 Fast. The exact cost is set by the model you pick.
Not here. Running CogVideoX on your own machine needs an NVIDIA GPU with about 5 GB of VRAM or more. On cogvideo.net the model runs on our servers, so any computer or phone with a browser works. If you would rather self-host, see the system requirements guide.
Short video clips from a text prompt, motion added to a single still image, and restyled versions of clips you already have. That covers social posts, ads, product animations, explainers, and quick concept demos.
Running it yourself means installing Python, CUDA, and the model weights, and owning a capable NVIDIA GPU. Here you skip all of that: open the page, type a prompt or upload an image, and generate. Self-hosting gives you full local control, while the hosted version gives you zero setup.
CogVideo is the original open research model from THUDM, and CogVideoX is the newer, stronger family that grew out of it, including releases like CogVideoX-2B and CogVideoX-5B. On this site, CogVideo is the hosted generator built on the CogVideoX line.
Right now you can generate with CogVideo v1, the CogVideoX-based generator, and with Veo 3.1 Fast. More models, such as Seedance, Kling, and others, are being added and will appear here as they go live.
Start Creating with CogVideoX
No GPU, no install. Generate your first video free, right in your browser, on the open source CogVideoX model.