bugfix

app.py

@@ -31,7 +31,7 @@ os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'max_split_size_mb:30'
 login(token=HF_TOKEN)
 
 MAX_SEED = np.iinfo(np.int32).max
-IMAGE_SIZE = 768
+IMAGE_SIZE = 1024
 
 # init
 device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -44,9 +44,7 @@ controlnet = FluxControlNetModel.from_pretrained(controlnet_model, torch_dtype=t
 pipe = FluxControlNetInpaintPipeline.from_pretrained(base_model, controlnet=controlnet, torch_dtype=torch.bfloat16).to(device)
 pipe.to("cuda")
 torch.backends.cuda.matmul.allow_tf32 = True
-pipe.vae.enable_tiling()
-pipe.vae.enable_slicing()
-pipe.enable_model_cpu_offload() # for saving memory
+
 
 control_mode_ids = {
     "scribble_hed": 0,
@@ -144,7 +142,7 @@ def upload_image_to_r2(image, account_id, access_key, secret_key, bucket_name):
     buffer.seek(0)
     s3.upload_fileobj(buffer, bucket_name, image_file)
     print("upload finish", image_file)
-    clear_cuda_cache()
+    
     return image_file
 
 
@@ -176,6 +174,10 @@ def run_flux(
         seed_slicer = random.randint(0, MAX_SEED)
     generator = torch.Generator().manual_seed(seed_slicer)
 
+    pipe.vae.enable_tiling()
+    pipe.vae.enable_slicing()
+    pipe.enable_model_cpu_offload() # for saving memory
+
     with calculateDuration("run pipe"):
         generated_image = pipe(
             prompt=prompt,