generate.py

import argparse
import subprocess
import os
import torch
from huggingface_hub import snapshot_download

# Arguments
parser = argparse.ArgumentParser()
parser.add_argument("--task", type=str, default="t2v-14B")
parser.add_argument("--size", type=str, default="832*480")
parser.add_argument("--frame_num", type=int, default=60)
parser.add_argument("--sample_steps", type=int, default=20)
parser.add_argument("--ckpt_dir", type=str, default="./Wan2.1-T2V-14B")
parser.add_argument("--offload_model", type=str, default="True")
parser.add_argument("--prompt", type=str, required=True)
args = parser.parse_args()

# Ensure the model is downloaded
if not os.path.exists(args.ckpt_dir):
    print("🔄 Downloading WAN 2.1 - 14B model from Hugging Face...")
    snapshot_download(repo_id="Wan-AI/Wan2.1-T2V-14B", local_dir=args.ckpt_dir)

# Free up GPU memory
if torch.cuda.is_available():
    torch.cuda.empty_cache()
    torch.backends.cudnn.benchmark = False
    torch.backends.cudnn.deterministic = True

# Run WAN 2.1 - 14B Model
command = f"python generate.py --task {args.task} --size {args.size} --frame_num {args.frame_num} --sample_steps {args.sample_steps} --ckpt_dir {args.ckpt_dir} --offload_model {args.offload_model} --prompt \"{args.prompt}\""

process = subprocess.Popen(command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
stdout, stderr = process.communicate()

# Print logs for debugging
print("🔹 Output:", stdout.decode())
print("🔺 Error:", stderr.decode())

# Verify if video was created
if os.path.exists("output.mp4"):
    print("✅ Video generated successfully: output.mp4")
else:
    print("❌ Error: Video file not found!")