Wan 2.1：16.1K+ Star

# Clone the repository
git clone https://github.com/Wan-Video/Wan2.1.git
cd Wan2.1

# Create a virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies (torch >= 2.4.0 required)
pip install -r requirements.txt

# Install dependencies
poetry install

# If flash-attn fails, use no-build-isolation
poetry run pip install --upgrade pip setuptools wheel
poetry run pip install flash-attn --no-build-isolation
poetry install

# Install huggingface-cli
pip install "huggingface_hub[cli]"

# Download the 14B text-to-video model
huggingface-cli download Wan-AI/Wan2.1-T2V-14B --local-dir ./Wan2.1-T2V-14B

# Download the 1.3B model for consumer GPUs
huggingface-cli download Wan-AI/Wan2.1-T2V-1.3B --local-dir ./Wan2.1-T2V-1.3B

# Download the VAE
huggingface-cli download Wan-AI/Wan2.1-VAE --local-dir ./Wan2.1-VAE

# Download the text encoder
huggingface-cli download Wan-AI/Wan2.1-T5 --local-dir ./Wan2.1-T5

pip install modelscope
modelscope download Wan-AI/Wan2.1-T2V-14B --local_dir ./Wan2.1-T2V-14B

python generate.py \
  --task t2v-1.3B \
  --size 832*480 \
  --ckpt_dir ./Wan2.1-T2V-1.3B \
  --prompt "A serene mountain lake at sunrise, mist rising from the water, camera slowly panning right"

python generate.py \
  --task t2v-14B \
  --size 1280*720 \
  --ckpt_dir ./Wan2.1-T2V-14B \
  --prompt "Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage."

cd gradio

# Run T2V 14B with single GPU
python t2v_14B_singleGPU.py \
  --prompt_extend_method 'dashscope' \
  --ckpt_dir ./Wan2.1-T2V-14B

# Run T2V 1.3B (lighter, for consumer GPUs)
python t2v_1.3B_singleGPU.py \
  --ckpt_dir ./Wan2.1-T2V-1.3B

# Install custom nodes
cd ComfyUI/custom_nodes
git clone https://github.com/Kijai/ComfyUI-WanVideoWrapper.git
git clone https://github.com/Kosinkadink/ComfyUI-VideoHelperSuite.git
git clone https://github.com/kijai/ComfyUI-KJNodes.git

# Install node dependencies
cd ComfyUI-WanVideoWrapper
pip install -r requirements.txt

# Diffusion models -> ComfyUI/models/diffusion_models
# Wan2_1-T2V-14B_fp8_e4m3fn.safetensors
# Wan2_1-T2V-1_3B_fp32.safetensors

# Text encoders -> ComfyUI/models/text_encoders
# umt5-xxl-enc-bf16.safetensors

# VAE -> ComfyUI/models/vae
# Wan2_1_VAE_fp32.safetensors

import torch
from diffusers.utils import export_to_video
from diffusers import AutoencoderKLWan, WanPipeline
from diffusers.schedulers.scheduling_unipc_multistep import UniPCMultistepScheduler

# Load model
model_id = "Wan-AI/Wan2.1-T2V-14B-Diffusers"
vae = AutoencoderKLWan.from_pretrained(
    model_id, subfolder="vae", torch_dtype=torch.float32
)

# Configure scheduler
flow_shift = 5.0  # 5.0 for 720P, 3.0 for 480P
scheduler = UniPCMultistepScheduler(
    prediction_type='flow_prediction',
    use_flow_sigmas=True,
    num_train_timesteps=1000,
    flow_shift=flow_shift
)

# Build pipeline
pipe = WanPipeline.from_pretrained(
    model_id, vae=vae, torch_dtype=torch.bfloat16
)
pipe.scheduler = scheduler
pipe.to("cuda")

# Generate
prompt = (
    "A cat and a dog baking a cake together in a kitchen. "
    "The cat is carefully measuring flour, while the dog is stirring "
    "the batter with a wooden spoon. The kitchen is cozy, with sunlight "
    "streaming through the window."
)

negative_prompt = (
    "Bright tones, overexposed, static, blurred details, subtitles, "
    "style, works, paintings, images, static, overall gray, worst quality, "
    "low quality, JPEG compression residue, ugly, incomplete"
)

output = pipe(
    prompt=prompt,
    negative_prompt=negative_prompt,
    height=720,
    width=1280,
    num_frames=81,
    guidance_scale=5.0,
).frames[0]

export_to_video(output, "output.mp4", fps=16)

# Install xDiT
pip install "xfuser>=0.4.1"

# Run on 8 GPUs
torchrun --nproc_per_node=8 generate.py \
  --task t2v-14B \
  --size 1280*720 \
  --ckpt_dir ./Wan2.1-T2V-14B \
  --dit_fsdp --t5_fsdp \
  --ulysses_size 8 \
  --prompt "Your prompt here"

python generate.py \
  --task i2v-14B \
  --size 1280*720 \
  --ckpt_dir ./Wan2.1-I2V-14B-720P \
  --image examples/i2v_input.JPG \
  --prompt "Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard."

python generate.py \
  --task flf2v-14B \
  --size 1280*720 \
  --ckpt_dir ./Wan2.1-FLF2V-14B-720P \
  --first_frame examples/flf2v_input_first_frame.png \
  --last_frame examples/flf2v_input_last_frame.png \
  --prompt "CG animation style, a small blue bird takes off from the ground, flapping its wings."

# Using local Qwen model
python generate.py \
  --task t2v-14B \
  --size 1280*720 \
  --ckpt_dir ./Wan2.1-T2V-14B \
  --prompt "A cat playing piano" \
  --use_prompt_extend \
  --prompt_extend_model Qwen/Qwen2.5-7B-Instruct

# Using DashScope API
DASH_API_KEY=your_key python generate.py \
  --task t2v-14B \
  --size 1280*720 \
  --ckpt_dir ./Wan2.1-T2V-14B \
  --prompt "A cat playing piano" \
  --use_prompt_extend \
  --prompt_extend_method 'dashscope'

# FP8 quantization reduces VRAM by ~20%
python generate.py \
  --task t2v-14B \
  --size 832*480 \
  --ckpt_dir ./Wan2.1-T2V-14B \
  --offload_model True \
  --t5_cpu \
  --prompt "Your prompt here"

FROM nvidia/cuda:12.1.0-devel-ubuntu22.04

WORKDIR /app
RUN apt-get update && apt-get install -y python3-pip git

COPY requirements.txt .
RUN pip install -r requirements.txt

COPY . .
RUN huggingface-cli download Wan-AI/Wan2.1-T2V-14B \
    --local-dir ./Wan2.1-T2V-14B

EXPOSE 7860
CMD ["python", "gradio/t2v_14B_singleGPU.py", "--ckpt_dir", "./Wan2.1-T2V-14B"]

docker build -t wan2.1 .
docker run --gpus all -p 7860:7860 wan2.1

import time
import psutil
import torch
from wan.utils.generation import generate_video

def generate_with_monitoring(prompt, **kwargs):
    process = psutil.Process()
    start_mem = process.memory_info().rss / 1024**3
    start_time = time.time()
    
    result = generate_video(prompt, **kwargs)
    
    elapsed = time.time() - start_time
    peak_mem = process.memory_info().rss / 1024**3
    gpu_mem = torch.cuda.max_memory_allocated() / 1024**3
    
    print(f"Generation completed in {elapsed:.1f}s")
    print(f"Peak GPU memory: {gpu_mem:.1f} GB")
    print(f"RAM delta: {peak_mem - start_mem:.1f} GB")
    
    return result

# Install DiffSynth-Studio
pip install diffsynth-studio

# Train a style LoRA
python -m diffsynth.train \
  --model_name wan2.1-t2v-14b \
  --dataset_path ./my_style_videos \
  --output_path ./wan_lora_output \
  --learning_rate 1e-4 \
  --num_train_steps 1000