Update app.py

app.py

@@ -20,7 +20,7 @@ from funcs import (
     get_latent_z,
     save_videos
 )
-from transformers import pipeline
+from transformers import pipeline, AutoTokenizer, AutoModelForSeq2SeqLT
 from diffusers import StableDiffusionXLPipeline
 
 print("PyTorch version:", torch.__version__)
@@ -29,6 +29,19 @@ print("CUDA available:", torch.cuda.is_available())
 def is_tensor(x):
     return torch.is_tensor(x)
 
+# 번역 모델 로드 (PyTorch 버전 사용)
+translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ko-en", device=0 if torch.cuda.is_available() else -1, framework="pt")
+
+# 이미지 생성 모델 로드
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+pipe = StableDiffusionXLPipeline.from_pretrained(
+    "SG161222/RealVisXL_V4.0",
+    torch_dtype=torch.float32,
+    use_safetensors=True,
+    add_watermarker=False
+).to(device)
+
+
 os.environ['KERAS_BACKEND'] = 'pytorch'
 
 def download_model():
@@ -56,17 +69,7 @@ model.eval()
 model = torch.nn.DataParallel(model)
 model = model.cuda()
 
-# 번역 모델 로드
-translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ko-en", device=0 if torch.cuda.is_available() else -1, framework="pt")
 
-# 이미지 생성 모델 로드
-device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
-pipe = StableDiffusionXLPipeline.from_pretrained(
-    "SG161222/RealVisXL_V4.0",
-    torch_dtype=torch.float32,
-    use_safetensors=True,
-    add_watermarker=False
-).to(device)
 
 def generate_image(prompt: str):
     # 한글 입력 감지 및 번역
@@ -158,12 +161,12 @@ def infer(prompt, steps=50, cfg_scale=7.5, eta=1.0, fs=3, seed=123, frames=64):
         torch.cuda.empty_cache()
         
         return video_path
+
     except Exception as e:
         print(f"Error occurred: {e}")
         return None
     finally:
-        torch.cuda.empty_cache()
-        
+        torch.cuda.empty_cache()        
 
 i2v_examples = [
     ['우주인 복장으로 기타를 치는 남자', 30, 7.5, 1.0, 6, 123, 64],