Create app.py

app.py

@@ -0,0 +1,23 @@
+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")