Update app.py

app.py

@@ -30,20 +30,26 @@ high_threshold = 200
 # Generator seed,
 generator = torch.manual_seed(0)
 
+torch_dtype = torch.float16  # or torch.bfloat16 (if needed)
+
 pose_model = OpenposeDetector.from_pretrained("lllyasviel/ControlNet")
+
 controlnet_pose = ControlNetModel.from_pretrained(
-    "lllyasviel/sd-controlnet-openpose", torch_dtype=torch.float16
-)
+    "lllyasviel/sd-controlnet-openpose", torch_dtype=torch_dtype
+).to("cuda")  # Load it directly on GPU
+
 pipe_pose = StableDiffusionControlNetPipeline.from_pretrained(
-    "runwayml/stable-diffusion-v1-5", controlnet=controlnet_pose, safety_checker=None, torch_dtype=torch.float16
+    "runwayml/stable-diffusion-v1-5",
+    controlnet=controlnet_pose,
+    safety_checker=None,
+    torch_dtype=torch_dtype
 )
+
 pipe_pose.scheduler = UniPCMultistepScheduler.from_config(pipe_pose.scheduler.config)
 
-# This command loads the individual model components on GPU on-demand. So, we don't
-# need to explicitly call pipe.to("cuda").
 pipe_pose.enable_model_cpu_offload()
 
-# xformers
+# ✅ Enable xformers (Optimizes memory usage)
 pipe_pose.enable_xformers_memory_efficient_attention()