app.py

import streamlit as st
import torch
from diffusers import AutoencoderKLWan, WanPipeline
from diffusers.utils import export_to_video

# Load the Wan2.1 text-to-video pipeline (1.3B version) with half precision weights
model_id = "Wan-AI/Wan2.1-T2V-1.3B-Diffusers"
st.write("Downloading and loading model... (first run may take a few minutes)")
vae = AutoencoderKLWan.from_pretrained(model_id, subfolder="vae", torch_dtype=torch.float16)
pipe = WanPipeline.from_pretrained(model_id, vae=vae, torch_dtype=torch.float16)
# (By default, the pipeline is on CPU since no .to("cuda") is called)

st.title("Wan2.1 Text-to-Video Generator")
prompt = st.text_input("Enter a text prompt for the video:")
frames = st.slider("Number of frames (video length)", min_value=8, max_value=81, value=24)
if st.button("Generate Video") and prompt:
    with st.spinner("Generating video... this may take a while on CPU"):
        # Run the pipeline to generate video frames
        result = pipe(prompt=prompt, height=480, width=832, num_frames=frames, num_inference_steps=20)
        video_frames = result.frames  # list of PIL images
        # Save frames as video file
        export_to_video(video_frames, "output.mp4", fps=8)  # using a lower FPS for a short video
    st.video("output.mp4")