Update app.py

app.py

@@ -18,6 +18,7 @@ def sh(cmd): subprocess.check_call(cmd, shell=True)
 
 flash_attention_installed = False
 
+# FlashAttention 설치 시도 - 실패해도 계속 진행
 try:
     print("Attempting to download and install FlashAttention wheel...")
     flash_attention_wheel = hf_hub_download(
@@ -38,6 +39,26 @@ except Exception as e:
     print(f"⚠️ Could not install FlashAttention: {e}")
     print("Continuing without FlashAttention...")
 
+# ===== CRITICAL FIX: attention.py 파일 패치 =====
+attention_file = "/home/user/app/ovi/modules/attention.py"
+if os.path.exists(attention_file):
+    try:
+        with open(attention_file, 'r') as f:
+            content = f.read()
+        
+        # FLASH_ATTN_3_AVAILABLE 변수가 정의되지 않은 경우를 위한 패치
+        if 'FLASH_ATTN_3_AVAILABLE' not in content.split('try:')[0]:
+            # 파일 시작 부분에 변수 초기화 추가
+            patched_content = f"FLASH_ATTN_3_AVAILABLE = False\n\n{content}"
+            
+            with open(attention_file, 'w') as f:
+                f.write(patched_content)
+            
+            print("✓ Successfully patched attention.py")
+    except Exception as e:
+        print(f"⚠️ Could not patch attention.py: {e}")
+# ===== END FIX =====
+
 import torch
 print(f"Torch version: {torch.__version__}")
 print(f"FlashAttention available: {flash_attention_installed}")
@@ -101,30 +122,26 @@ snapshot_download(
 # Initialize OviFusionEngine
 enable_cpu_offload = args.cpu_offload
 print(f"loading model...")
-DEFAULT_CONFIG['cpu_offload'] = enable_cpu_offload # always use cpu offload if image generation is enabled
-DEFAULT_CONFIG['mode'] = "t2v"  # hardcoded since it is always cpu offloaded
+DEFAULT_CONFIG['cpu_offload'] = enable_cpu_offload
+DEFAULT_CONFIG['mode'] = "t2v"
 ovi_engine = OviFusionEngine()
 print("loaded model")
 
 
 def resize_for_model(image_path):
-    # Open image
     img = Image.open(image_path)
     w, h = img.size
     aspect_ratio = w / h
     
-    # Decide target size based on aspect ratio
-    if aspect_ratio > 1.5:  # wide image
+    if aspect_ratio > 1.5:
         target_size = (992, 512)
-    elif aspect_ratio < 0.66:  # tall image
+    elif aspect_ratio < 0.66:
         target_size = (512, 992)
-    else:  # roughly square
+    else:
         target_size = (512, 512)
 
-    # Resize while preserving aspect ratio, then pad
     img.thumbnail(target_size, Image.Resampling.LANCZOS)
 
-    # Create a new image with target size and paste centered
     new_img = Image.new("RGB", target_size, (0, 0, 0))
     new_img.paste(
         img,
@@ -153,11 +170,9 @@ def generate_scene(
     if not image:
         raise gr.Error("Please provide an image")
     
-
     if not text_prompt_processed:
         raise gr.Error("Please enter a prompt.")
     
-    
     return generate_video(text_prompt, 
                         image, 
                         sample_steps, 
@@ -188,7 +203,6 @@ def get_duration(
     progress,
 ):
     warmup = 20
-
     return int(sample_steps * 3 + warmup)
     
 
@@ -217,12 +231,10 @@ def generate_video(
         if session_id is None:
             session_id = uuid.uuid4().hex
             
-
         output_dir = os.path.join(os.environ["PROCESSED_RESULTS"], session_id)
         os.makedirs(output_dir, exist_ok=True)
         output_path = os.path.join(output_dir, f"generated_video.mp4")
 
-
         _, target_size = resize_for_model(image_path)
 
         video_frame_width = target_size[0]
@@ -252,14 +264,12 @@ def generate_video(
 
 
 def cleanup(request: gr.Request):
-
     sid = request.session_hash
     if sid:
         d1 = os.path.join(os.environ["PROCESSED_RESULTS"], sid)
         shutil.rmtree(d1, ignore_errors=True)
         
 def start_session(request: gr.Request):
-
     return request.session_hash
 
 css = """
@@ -297,7 +307,6 @@ with gr.Blocks(css=css, theme=theme) as demo:
         )
         with gr.Row():
             with gr.Column():
-                # Image section
                 image = gr.Image(type="filepath", label="Image", height=360)
 
                 video_text_prompt = gr.Textbox(label="Scene Prompt", 
@@ -341,37 +350,31 @@ with gr.Blocks(css=css, theme=theme) as demo:
                     video_negative_prompt = gr.Textbox(label="Video Negative Prompt", placeholder="Things to avoid in video")
                     audio_negative_prompt = gr.Textbox(label="Audio Negative Prompt", placeholder="Things to avoid in audio")
 
-
             with gr.Column():
                 output_path = gr.Video(label="Generated Video", height=360)
 
                 gr.Examples(
                     examples=[
-
                         [
                             "The video opens with a close-up of a woman with vibrant reddish-orange, shoulder-length hair and heavy dark eye makeup. She is wearing a dark brown leather jacket over a grey hooded top. She looks intently to her right, her mouth slightly agape, and her expression is serious and focused. The background shows a room with light green walls and dark wooden cabinets on the left, and a green plant on the right. She speaks, her voice clear and direct, saying, <S>doing<E>. She then pauses briefly, her gaze unwavering, and continues, <S>And I need you to trust them.<E>. Her mouth remains slightly open, indicating she is either about to speak more or has just finished a sentence, with a look of intense sincerity.. <AUDCAP>Tense, dramatic background music, clear female voice.<ENDAUDCAP>", 
                             "example_prompts/pngs/8.png",
                             50,
                         ],
-
                         [
                             "A young woman with long, wavy blonde hair and light-colored eyes is shown in a medium shot against a blurred backdrop of lush green foliage. She wears a denim jacket over a striped top. Initially, her eyes are closed and her mouth is slightly open as she speaks, <S>Enjoy this moment<E>. Her eyes then slowly open, looking slightly upwards and to the right, as her expression shifts to one of thoughtful contemplation. She continues to speak, <S>No matter where it's taking<E>, her gaze then settling with a serious and focused look towards someone off-screen to her right.. <AUDCAP>Clear female voice, faint ambient outdoor sounds.<ENDAUDCAP>",
                             "example_prompts/pngs/2.png",
                             50,
                         ],
-
                         [
                             "A bearded man wearing large dark sunglasses and a blue patterned cardigan sits in a studio, actively speaking into a large, suspended microphone. He has headphones on and gestures with his hands, displaying rings on his fingers. Behind him, a wall is covered with red, textured sound-dampening foam on the left, and a white banner on the right features the ""CHOICE FM"" logo and various social media handles like ""@ilovechoicefm"" with ""RALEIGH"" below it. The man intently addresses the microphone, articulating, <S>is talent. It's all about authenticity. You gotta be who you really are, especially if you're working<E>. He leans forward slightly as he speaks, maintaining a serious expression behind his sunglasses.. <AUDCAP>Clear male voice speaking into a microphone, a low background hum.<ENDAUDCAP>",
                             "example_prompts/pngs/5.png",
                             50,
                         ],
-
                         [
                             "The video opens with a close-up on an older man with long, grey hair and a short, grey beard, wearing dark sunglasses. He is clad in a dark coat, possibly with fur trim, and black gloves. His face is angled slightly upwards and to the right, as he begins to speak, his mouth slightly open. In the immediate foreground, out of focus, is the dark-clad shoulder and the back of the head of another person. The man articulates, <S>labbra. Ti ci vorrebbe...<E> His expression remains contemplative, and he continues, seemingly completing his thought, <S>Un ego solare.<E> The background behind him is a textured, grey stone wall, suggesting an outdoor setting. The man's gaze remains fixed upwards, his expression thoughtful.. <AUDCAP>A clear, slightly low-pitched male voice speaking Italian. The overall soundscape is quiet, with no prominent background noises or music.<ENDAUDCAP>",
                             "example_prompts/pngs/7.png",
                             50,
                         ],
-
                         [
                             "The scene is set outdoors with a blurry, bright green background, suggesting grass and a sunny environment. On the left, a woman with long, dark hair, wearing a red top and a necklace with a white pendant, faces towards the right. Her expression is serious and slightly perturbed as she speaks, with her lips slightly pursed. She says, <S>UFO, UFC thing.<E> On the right, the back of a man's head and his right ear are visible, indicating he is facing away from the camera, listening to the woman. He has short, dark hair. The woman continues speaking, her expression remaining serious, <S>And if you're not watching that, it's one of those ancient movies from an era that's<E> as the frame holds steady on the two figures.. <AUDCAP>Clear female speech, distant low-frequency hum.<ENDAUDCAP>",
                             "example_prompts/pngs/9.png",
@@ -393,5 +396,4 @@ with gr.Blocks(css=css, theme=theme) as demo:
 if __name__ == "__main__":
     demo.unload(cleanup)
     demo.queue()
-    demo.launch(ssr_mode=False, share=True)
-
+    demo.launch(ssr_mode=False, share=True)