Update app.py

app.py

@@ -17,13 +17,18 @@ base_model_repo = "stabilityai/stable-diffusion-3-medium-diffusers"
 lora_weights_path = "./pytorch_lora_weights.safetensors"
 
 # Load the base model
-pipeline = DiffusionPipeline.from_pretrained(base_model_repo, torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32, use_auth_token=HUGGINGFACE_TOKEN)
+pipeline = DiffusionPipeline.from_pretrained(
+    base_model_repo,
+    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+    use_auth_token=HUGGINGFACE_TOKEN
+)
 pipeline.load_lora_weights(lora_weights_path)
-pipeline.enable_sequential_cpu_offload()
+pipeline.enable_sequential_cpu_offload()  # Efficient memory usage
+pipeline.enable_xformers_memory_efficient_attention()  # Enable xformers memory efficient attention
 pipeline = pipeline.to(device)
 
 MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
+MAX_IMAGE_SIZE = 768  # Reduce max image size to fit within memory constraints
 
 def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps):
     if randomize_seed:
@@ -128,9 +133,9 @@ with gr.Blocks(css=css) as demo:
                 num_inference_steps = gr.Slider(
                     label="Number of inference steps",
                     minimum=1,
-                    maximum=100,
+                    maximum=50,
                     step=1,
-                    value=50,
+                    value=30,
                 )
         
         gr.Examples(