Update app.py

app.py

@@ -9,6 +9,8 @@ import torch
 from facenet_pytorch import MTCNN, InceptionResnetV1
 import numpy as np
 from PIL import Image
+import base64
+import tempfile
 from transformers import AutoProcessor, LlavaOnevisionForConditionalGeneration
 
 import cv2
@@ -141,8 +143,17 @@ def scenes_extraction(video_file: str, threshold: float, offset_frames: int, cro
     # video_file es un str ya que aunque realmente el usuario subió un archivo desde la UI, Gradio lo guarda temporalmente como ruta
     
     try:
+        video_bytes = base64.b64decode(video_file)
+
+        # archivo temporal en /tmp
+        temp_video = tempfile.NamedTemporaryFile(delete=False, suffix=".mp4")
+        temp_video.write(video_bytes)
+        temp_video.flush()
+        temp_video.close()
+
+        video_path = temp_video.name
         # Detectamos las escenas
-        video_manager = VideoManager([video_file])
+        video_manager = VideoManager([video_path])
         scene_manager = SceneManager()
         scene_manager.add_detector(ContentDetector(threshold=threshold))
         video_manager.start()
@@ -231,7 +242,7 @@ with gr.Blocks(title="Salamandra Vision 7B · ZeroGPU") as demo:
     face_btn.click(face_image_embedding, [face_img], face_out, api_name="face_image_embedding", concurrency_limit=1)
 
     with gr.Row():
-        video_file = gr.Video(label="Sube un vídeo")
+        video_file = gr.Textbox(label="Texto/prompt", value="Base64 del video")
         threshold = gr.Slider(0.0, 100.0, value=30.0, step=1.0, label="Threshold")
         offset_frames = gr.Slider(0, 30, value=5, step=1, label="Offset frames")
         crop_ratio = gr.Slider(0.0, 1.0, value=1.0, step=0.05, label="Crop ratio")