Wan2.1 (Wan-Video/Wan2.1)

Licence: Apache-2.0. Cloned shallow for review at /home/user/wan-video/wan2.1 (re-clone with git clone --depth 1 https://github.com/Wan-Video/Wan2.1 if the path is missing). Weights are not in the repo.

What it is

Open suite of video models: T2V (1.3B and 14B), I2V-14B (480P/720P), FLF2V-14B (first and last frame), VACE (all-in-one creation/editing), plus Wan-VAE. Also integrated in ComfyUI and Diffusers. README claims the T2V-1.3B model needs about 8.19 GB VRAM and makes a 5-second 480P clip on an RTX 4090 in roughly 4 minutes.

Setup

git clone https://github.com/Wan-Video/Wan2.1 && cd Wan2.1
pip install -r requirements.txt          # check INSTALL.md for torch/flash-attn pins
pip install "huggingface_hub[cli]"
huggingface-cli download Wan-AI/Wan2.1-T2V-1.3B --local-dir ./Wan2.1-T2V-1.3B

Generate

# small GPU (480P)
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 "..."
# 720P text-to-video
python generate.py --task t2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-T2V-14B --prompt "..."
# image-to-video
python generate.py --task i2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-I2V-14B-720P --image in.jpg --prompt "..."
# first/last frame
python generate.py --task flf2v-14B --size 1280*720 --ckpt_dir ./Wan2.1-FLF2V-14B-720P --first_frame a.png --last_frame b.png --prompt "..."

Multi-GPU: torchrun --nproc_per_node=8 generate.py ... --dit_fsdp --t5_fsdp --ulysses_size 8 (needs xfuser>=0.4.1). Prompt extension: --use_prompt_extend with a local Qwen model or DashScope (needs DASH_API_KEY).

Guidance

• Pick 1.3B for a single consumer GPU, 14B only with large VRAM or multi-GPU.
• Prompt structure: ai-video-image-prompt-structure.
• Check the Hugging Face model card licence before commercial use; likeness and voice rules in ai-video-generators apply.

Related

ai-video-generators, repo-hunyuanvideo, repo-cogvideo, repo-pinokio (one-click local launcher).