Update app.py

app.py

@@ -14,13 +14,7 @@ from torch.nn.utils.parametrizations import weight_norm
 
 login(token=os.environ["HF_TOKEN"])
 
-
 device = torch.device("cpu")
-processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
-model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large").to("cpu")
-pipe = StableAudioPipeline.from_pretrained("stabilityai/stable-audio-open-1.0")
-pipe = pipe.to("cpu")
-
 
 #img_url = 'https://www.caracteristicass.de/wp-content/uploads/2023/02/imagenes-artisticas.jpg'
 
@@ -38,6 +32,8 @@ with gr.Blocks(theme=gr.themes.Ocean(primary_hue="pink", neutral_hue="indigo", f
     output2 = gr.Audio(label="Audio")
 
     def describir(url):
+      processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
+      model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large").to("cpu")
       raw_image = Image.open(requests.get(url, stream=True).raw).convert('RGB')
       inputs = processor(raw_image, return_tensors="pt").to("cpu")
       out = model.generate(**inputs)
@@ -45,6 +41,8 @@ with gr.Blocks(theme=gr.themes.Ocean(primary_hue="pink", neutral_hue="indigo", f
       return leer(processor.decode(out[0], skip_special_tokens=True))
 
     def leer(texto):
+        pipe = StableAudioPipeline.from_pretrained("stabilityai/stable-audio-open-1.0")
+        pipe = pipe.to("cpu")
         prompt = texto
         negative_prompt = "Low quality."