Update app.py

app.py

@@ -30,6 +30,8 @@ cuda_device = 'cuda' if torch.cuda.is_available() else 'cpu'
 sr_model = RealESRGAN(torch.device('cuda' if torch.cuda.is_available() else 'cpu'), scale=2)
 sr_model.load_weights(f'weights/RealESRGAN_x2.pth', download=False)
 
+last_uploaded_image = None
+
 @torch.inference_mode()
 @spaces.GPU(duration=120) 
 def multimodal_understanding(image, question, seed, top_p, temperature, progress=gr.Progress(track_tqdm=True)):
@@ -76,12 +78,13 @@ with gr.Blocks(css=css) as demo:
     image_input = gr.Image(label="Upload an image (optional)")
     
     def respond(message, image):
-        # Here you can add logic to handle the image if provided
+        global last_uploaded_image
         if image is not None:
-            # Call multimodal understanding with the image and message
+            last_uploaded_image = image  # Update the last uploaded image
             response = multimodal_understanding(image, message, seed=42, top_p=0.95, temperature=0.1)
+        elif last_uploaded_image is not None:
+            response = multimodal_understanding(last_uploaded_image, message, seed=42, top_p=0.95, temperature=0.1)
         else:
-            # If no image is provided, just respond with a text-based answer
             response = "Please provide an image for multimodal understanding."
         
         return response