Update app.py

app.py

@@ -632,14 +632,18 @@ def preprocess_img(input_image_path, raw_image_path, session_id = None):
 
 def infer_example(image_path, audio_path, num_steps, raw_image_path, session_id = None, progress=gr.Progress(track_tqdm=True),):
 
+    current_image_size = args.image_sizes_720
+    args.image_sizes_720 = [[720, 400]]
     text = ADAPTIVE_PROMPT_TEMPLATES[2]
 
-    result = infer(image_path, audio_path, text, num_steps, [[720, 400]], session_id, progress)
+    result = infer(image_path, audio_path, text, num_steps, session_id, progress)
 
-    return result, image_ratio
+    args.image_sizes_720 = current_image_size
+
+    return result
     
 @spaces.GPU(duration=get_duration)
-def infer(image_path, audio_path, text, num_steps, image_ratio, session_id = None, progress=gr.Progress(track_tqdm=True),):
+def infer(image_path, audio_path, text, num_steps, session_id = None, progress=gr.Progress(track_tqdm=True),):
 
     if image_path is None:
 
@@ -692,8 +696,7 @@ def infer(image_path, audio_path, text, num_steps, image_ratio, session_id = Non
 
     tmp2_audio_path = os.path.join(audio_dir, f"audio_out.wav")
     prompt_path = os.path.join(prompt_dir, f"prompt.txt") 
-
-    args.image_sizes_720 = image_ratio
+    
     video = inferpipe(
                 prompt=text,
                 image_path=clay_image_path,
@@ -739,15 +742,13 @@ def orientation_changed(session_id, evt: gr.EventData):
     detail = getattr(evt, "data", None) or getattr(evt, "_data", {}) or {}
 
     if detail['value'] == "9:16":
-        image_ratio = [[720, 400]]
+        args.image_sizes_720 = [[720, 400]]
     elif detail['value'] == "1:1":
-        image_ratio = [[720, 720]]
+        args.image_sizes_720 = [[720, 720]]
     elif detail['value'] == "16:9":
-        image_ratio = [[400, 720]]
-
-    print(f'{session_id} has {image_ratio} orientation')
+        args.image_sizes_720 = [[400, 720]]
 
-    return image_ratio
+    print(f'{session_id} has {args.image_sizes_720} orientation')
 
 def clear_raw_image():
     return ''
@@ -813,7 +814,6 @@ css = """
 with gr.Blocks(css=css) as demo:
 
     session_state = gr.State()
-    image_ratio = gr.State([[720, 400]])
     demo.load(start_session, outputs=[session_state])
 
 
@@ -911,7 +911,7 @@ with gr.Blocks(css=css) as demo:
                     ],
                     label="Cached Examples",
                     inputs=[image_input, audio_input, num_steps, raw_img_text],
-                    outputs=[output_video, image_ratio],
+                    outputs=[output_video],
                     fn=infer_example,
                     cache_examples=True
                     )
@@ -919,11 +919,11 @@ with gr.Blocks(css=css) as demo:
 
     infer_btn.click(
         fn=infer,
-        inputs=[image_input, audio_input, text_input, num_steps, image_ratio, session_state],
+        inputs=[image_input, audio_input, text_input, num_steps, session_state],
         outputs=[output_video]
     )
 
-    image_input.orientation(fn=orientation_changed, inputs=[session_state], outputs=[image_ratio]).then(fn=preprocess_img, inputs=[image_input, raw_img_text, session_state], outputs=[image_input, raw_img_text])
+    image_input.orientation(fn=orientation_changed, inputs=[session_state]).then(fn=preprocess_img, inputs=[image_input, raw_img_text, session_state], outputs=[image_input, raw_img_text])
     image_input.clear(fn=clear_raw_image, outputs=[raw_img_text])
     image_input.upload(fn=preprocess_img, inputs=[image_input, raw_img_text, session_state], outputs=[image_input, raw_img_text])
     image_input.change(fn=update_generate_button, inputs=[image_input, audio_input, text_input, num_steps, session_state], outputs=[time_required])