Video / Intermediate

Generate a short video on your own GPU

Run the publisher’s Wan2.1 1.3B text-to-video path at its recommended 480p setting and check the rendered clip.

Verified sourceSource checked 9/23/2026
LOCALRENTED GPU

Before you begin

Difficulty
Intermediate
Software
Python, PyTorch, Wan2.1
Hardware
The publisher gives an RTX 4090 offload example for the 1.3B checkpoint. Other GPUs need a separate memory and runtime check.
View the setup source

Sources and files

Publisher installation and generation commands Wan2.1 1.3B weights

Choose a model for this task

The steps below use the recommended model. Alternatives have their own package and command; open their model pages before switching.

The workflow

1

Prepare Python and a supported GPU

Use a Python environment with PyTorch 2.4 or later and a compatible CUDA GPU. The publisher’s single-GPU 1.3B example uses model and text-encoder offload on an RTX 4090.

2

Get the official inference code

Clone Wan’s repository and work inside the new directory.

git clone https://github.com/Wan-Video/Wan2.1.git
3

Install the repository dependencies

From inside the Wan2.1 directory install the requirements listed by the publisher.

pip install -r requirements.txt
4

Download the 1.3B checkpoint

Install the Hugging Face CLI, then download the Wan-AI/Wan2.1-T2V-1.3B repository into ./Wan2.1-T2V-1.3B. The publisher’s README links the correct checkpoint.

5

Generate a 480p test clip

Use the publisher’s 1.3B flags for 832×480, CPU text-encoder offload, and guide scale 6. Replace the example prompt after the first successful run.

python generate.py --task t2v-1.3B --size 832*480 --ckpt_dir ./Wan2.1-T2V-1.3B --offload_model True --t5_cpu --sample_shift 8 --sample_guide_scale 6 --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage."
6

Check the video result

Confirm the command completes and produces a playable video clip. Stay at 480p until you know the memory and quality behavior of your machine; the publisher recommends 480p for this 1.3B model.

When it doesn’t go to plan

The generation process exhausts GPU memory

Keep --offload_model True and --t5_cpu enabled, close other GPU apps, and start at the official 832×480 example size.

The checkpoint path is not found

Confirm the downloaded repository is inside Wan2.1-T2V-1.3B under the working directory and that --ckpt_dir matches that folder.

The model behind this workflow

Wan2.1 · 1.3B
MY HARDWARE

Will it run on your machine?

Save your machine to see a personalized rating and its reasoning.

Add my hardware

Keep exploring