Update app.py

app.py

@@ -9,6 +9,7 @@ from diffusers import Flux2Pipeline, Flux2Transformer2DModel
 import requests
 from PIL import Image
 import base64
+from gradio_client import Client
 
 dtype = torch.bfloat16
 device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -16,15 +17,15 @@ device = "cuda" if torch.cuda.is_available() else "cpu"
 MAX_SEED = np.iinfo(np.int32).max
 MAX_IMAGE_SIZE = 1024
 
+# Initialize text encoder client ONCE at module level to avoid thread exhaustion
+text_encoder_client = Client("multimodalart/mistral-text-encoder")
+
 def remote_text_encoder(prompts):
-    from gradio_client import Client
-    
-    client = Client("multimodalart/mistral-text-encoder")
-    result = client.predict(
+    result = text_encoder_client.predict(
         prompt=prompts,
         api_name="/encode_text"
     )
-    
+
     prompt_embeds = torch.load(result[0])
     return prompt_embeds