Update app.py

app.py

@@ -29,21 +29,27 @@ os.environ['TMPDIR'] = path_dict['gradio_temp_dir']
 
 
 # Function of loading pre-trained models.
-
-base_model_path = snapshot_download("black-forest-labs/FLUX.1-Fill-dev")
-image_encoder_path = snapshot_download("google/siglip-so400m-patch14-384")
-calligrapher_path = snapshot_download("Calligrapher2025/Calligrapher")
-transformer = FluxTransformer2DModel.from_pretrained(base_model_path, subfolder="transformer",
-                                                     torch_dtype=torch.bfloat16)
-pipe = CalligrapherPipeline.from_pretrained(base_model_path, transformer=transformer,
-                                            torch_dtype=torch.bfloat16).to("cuda")
-model = Calligrapher(pipe, image_encoder_path, calligrapher_path, device="cuda", num_tokens=128)
-
+def load_models():
+    base_model_path = snapshot_download("black-forest-labs/FLUX.1-Fill-dev")
+    image_encoder_path = snapshot_download("google/siglip-so400m-patch14-384")
+    calligrapher_path = snapshot_download("Calligrapher2025/Calligrapher")
+    transformer = FluxTransformer2DModel.from_pretrained(base_model_path, subfolder="transformer",
+                                                         torch_dtype=torch.bfloat16)
+    pipe = CalligrapherPipeline.from_pretrained(base_model_path, transformer=transformer,
+                                                torch_dtype=torch.bfloat16).to("cuda")
+    model = Calligrapher(pipe, image_encoder_path, calligrapher_path, device="cuda", num_tokens=128)
+
+    return model
+
+@gr.CachedFunction()
+def load_model_cached():
+    return load_models()
 
 @spaces.GPU
 def process_and_generate(editor_component, reference_image, prompt, height, width,
                          scale, steps=50, seed=42, use_context=True, num_images=1):
     print('Begin processing!')
+    model = load_model_cached()
     # Job directory.
     job_name = datetime.now().strftime("%Y-%m-%d-%H-%M-%S")
     job_dir = os.path.join(SAVE_DIR, job_name)