How do I install Wan2GP?
It is three commands. Total install time on a fresh box is about 15 minutes plus the model weight download.
bash
git clone https://github.com/deepbeepmeep/Wan2GP.git
cd Wan2GP
Step 1: Clone the repo
The repo is github.com/deepbeepmeep/Wan2GP — Apache 2.0 licensed, 5.6k stars, last release v11.52 "A Kind of Magic" on May 2, 2026.
Step 2: Run the one-click installer
bash
# Windows
install.bat
# Linux / Mac
bash install.sh
No CUDA toolkit setup, no Python venv wrangling. The installer pulls the right PyTorch build for your card automatically. First run downloads ~12GB of model weights — go make a coffee.
Step 3: Launch the web UI
bash
python app.py
Open http://127.0.0.1:7860 in your browser. You will see the Wan2GP interface with model selection at the top.
Step 4: Pick your model for the task
| Goal | Model to select | VRAM needed |
|---|---|---|
| General text-to-video | Wan 2.1 / 2.2 | 6 GB |
| Fast, stylized clips | LTX-2 | 8 GB |
| Cinematic quality | Hunyuan Video | 12 GB |
| Image + text prompt | Flux + Qwen Image | 10 GB |
| Talking-head avatars | Magi Human | 16 GB (1080p) |
Start with Wan 2.1 if you are unsure — it is the most tested option on a 6GB card.
Step 5: Write your first prompt
Be specific about motion, lighting, and camera angle. Try this:
A red fox running through a snowy forest at dusk, slow motion, cinematic lens flare, 4K
Set resolution to 480p first to validate, then bump to 720p once the motion is right. Hit Generate.
Step 6: Export and own your output
Videos save to /outputs in the project directory. No watermark, no usage clause, no cloud dependency. Drag them into any editor or post anywhere.
What is Wan2GP?
Wan2GP is a single one-click installer that wraps the top open-source video models — Wan 2.1/2.2, LTX-2, Hunyuan, Flux, Qwen Image, LongCat, Z-Image, Qwen3-TTS, Chatterbox — into one local web UI. The README sells itself in one line: "The best Open Source Video Generative Models Accessible to the GPU Poor". It supports legacy hardware including RTX 10-series and 20-series cards, plus AMD GPUs (RDNA 2-4).
The project hit 5.6k stars on GitHub by May 2026 and is actively maintained — release v11.52 dropped May 2, 2026.
Is Wan2GP really free?
Yes. The whole stack is Apache 2.0 licensed — including commercial use. No account required, no API key, no metered tokens. You pay for electricity and your own GPU. That is it.
The output files live in your /outputs folder with full ownership. Compare that with Runway's $12/month Standard plan which gives you 625 credits and stores your videos on their servers, or Pro at $28/month for 2,250 credits. Burn through credits and you are out of work until the next billing cycle.
How is Wan2GP different from Runway?
Three big differences:
- Hardware: Wan2GP runs on as little as 6GB of VRAM thanks to aggressive layer offloading. Runway runs nothing locally — every generation hits their cloud queue.
- Cost per clip: Wan2GP is free per generation. Runway's Standard at $12/month gets you 625 credits, which translates to roughly 60-100 short clips depending on resolution. Past that, you pay overage or wait.
- Ownership: Wan2GP outputs live on your disk under Apache 2.0. Runway clips live on their servers under their terms of service.
The README puts it directly: "Low VRAM requirements... Support for old Nvidia GPUs... Easy to use Full Web based interface." That is the whole pitch.
What's the catch?
Three honest caveats:
- Generation speed scales with your hardware. A 6GB card will be slower than Runway's cloud GPUs. If you need 50 clips today and you are on a laptop, the cloud is faster.
- First-time setup downloads ~12GB. If you are on metered internet, plan ahead.
- The UI is functional, not beautiful. It is a Gradio interface, not a polished SaaS product. You will trade design polish for not paying $76 a month.
Bottom line
If you have any GPU made in the last seven years, you should not be paying for AI video. Wan2GP runs the same model class as Runway's premium tools, locally, free, with full ownership. The only reason to pay for Runway is if you genuinely need their cloud infrastructure for a team workflow — and most people do not.
Related Nanoflow guides
Want a hand?
Book a 30-min call.
Walk through your stack with us. We'll find the bottleneck and map out the exact wiring you need — free.