Update app.py

app.py

@@ -1,27 +1,32 @@
 import streamlit as st
+from diffusers import DiffusionPipeline
+import torch
 from moviepy.editor import *
+import numpy as np
 import tempfile, os
-from transformers import pipeline
 
-st.title("Simplified Text-to-Video (No TTS temporarily)")
+st.title("🚀 Text-to-Video (Zeroscope)")
 
-@st.cache_resource()
+@st.cache_resource
 def load_model():
-    return pipeline('text-to-video-generation', model='cerspense/zeroscope_v2_576w')
+    pipe = DiffusionPipeline.from_pretrained(
+        "cerspense/zeroscope_v2_576w", 
+        torch_dtype=torch.float16
+    )
+    pipe.enable_cpu_offload()
+    return pipe
 
-video_gen = load_model()
+pipe = load_model()
 
-input_text = st.text_area("Enter short text (max 50 chars):", max_chars=50)
+prompt = st.text_area("Enter prompt (short & descriptive):", max_chars=50)
 
 if st.button("Generate Video"):
-    if input_text:
-        with st.spinner("Generating video..."):
-            video_output = video_gen(input_text, num_frames=10)
-            video_tensor = video_output["video"]
-            video_np = (video_tensor * 255).astype('uint8')
+    if prompt:
+        with st.spinner("Generating... (this may take ~2-3 mins)"):
+            video_frames = pipe(prompt, num_frames=10, height=320, width=576).frames
 
             video_filename = tempfile.mktemp(".mp4")
-            clips = [ImageClip(frame).set_duration(0.3) for frame in video_np]
+            clips = [ImageClip(np.array(frame)).set_duration(0.3) for frame in video_frames]
             final_clip = concatenate_videoclips(clips, method="compose")
             final_clip.write_videofile(video_filename, fps=5)
 
@@ -29,4 +34,4 @@ if st.button("Generate Video"):
 
             os.remove(video_filename)
     else:
-        st.warning("Please enter text.")
+        st.warning("Enter a prompt to generate a video.")