CogVideoX is a free, open-source text-to-video model from Zhipu AI that generates smooth, high-quality 5-second clips from a simple text prompt — and you can run it entirely on your own machine, no cloud and no censorship. It isn't as polished as closed models like Sora, Kling, or Luma, but it was a major leap for open-source local video. Best of all, the install is genuinely one click thanks to Pinokio.
The CogVideoX models: 2B vs 5B
CogVideoX comes in two sizes, each suited to different hardware and licensing needs:
| Model | Best for | Full-precision VRAM | License |
|---|---|---|---|
| CogVideoX-2B | Entry-level — easy to run, great starting point | ~18 GB | Apache 2.0 (very permissive) |
| CogVideoX-5B | Best quality, more detailed motion | ~26 GB | CogVideoX license (more restrictive) |
Why Pinokio? One-click local AI
Pinokio is a free application that installs and runs open-source AI tools for you — think of it as a local program runner that handles all the messy dependencies automatically. It has a one-click installer for CogVideoX, which is by far the easiest way to get it running on Windows.
Step-by-step: install CogVideoX with Pinokio
- Download Pinokio. Go to pinokio.computer, scroll to the download links, and grab the build for your OS (Windows, Mac, or Linux).
- Install and open Pinokio on your device.
- Open Discover. Click the Discover button in the top-right corner to see the full list of one-click open-source projects.
- Select CogVideo. It's one of the first options — click it, then click the Download button on its page.
- Install dependencies. You may get a popup about missing dependencies — just click Install at the bottom and Pinokio installs everything automatically. This can take a few minutes.
- Download the CogVideo repo. Once dependencies are in, a new download button appears for the CogVideo GitHub repo. Click it and let it download.
- Install the project. After the repo downloads, click the Install button that appears to install the project dependencies (another minute or two).
- Start the UI. If you see an error after the dependencies finish, hit Stop in the left sidebar, then Start — this relaunches the server and opens the CogVideo Gradio web UI right inside Pinokio.
Generating your first video
- Write a prompt in the prompt box. The model was trained on longer prompts, so more detail generally means better results.
- (Optional) Enhance prompt. The "enhance prompt" feature uses an OpenAI GPT model to expand your prompt — add your OpenAI API key in the Configure tab to enable it.
- Pick your model. Choose 5B or 2B, and the bfloat16 or float16 version (float16 is lighter on VRAM).
- Generate. Click Generate video. The first run downloads the model (12 GB+), so it takes longer. After that, generation typically takes 15 minutes or more depending on your specs — the model renders each frame individually.
- Watch it. When it's done, hit play to preview your clip right in the browser.
Looking for faster, newer local video?
CogVideoX was an early open-source breakthrough, but local video has moved fast since. If you want sharper output and faster generation on low VRAM today, check out these one-click setups:
- WAN 2.2 14B (Text/Image to Video) — runs on 6 GB VRAM via GGUF.
- LTX-2 via Wan2GP — fast low-VRAM video (see also our Wan2GP guide).
- Not sure what your GPU can handle? See the Local AI VRAM Requirements 2026 guide.