Add optimizations

app.py

@@ -8,9 +8,15 @@ import json
 
 from PIL import Image
 from diffusers import QwenImageEditPipeline, FlowMatchEulerDiscreteScheduler
+
 from huggingface_hub import InferenceClient
 import math
 
+from optimization import optimize_pipeline_
+from qwenimage.pipeline_qwen_image_edit import QwenImageEditPipeline as QwenImageEditPipelineCustom
+from qwenimage.transformer_qwenimage import QwenImageTransformer2DModel
+from qwenimage.qwen_fa3_processor import QwenDoubleStreamAttnProcessorFA3
+
 # --- Prompt Enhancement using Hugging Face InferenceClient ---
 def polish_prompt_hf(original_prompt, system_prompt):
     """
@@ -159,7 +165,7 @@ scheduler_config = {
 scheduler = FlowMatchEulerDiscreteScheduler.from_config(scheduler_config)
 
 # Load the edit pipeline with Lightning scheduler
-pipe = QwenImageEditPipeline.from_pretrained(
+pipe = QwenImageEditPipelineCustom.from_pretrained(
     "Qwen/Qwen-Image-Edit", 
     scheduler=scheduler,
     torch_dtype=dtype
@@ -177,6 +183,13 @@ except Exception as e:
     print(f"Warning: Could not load Lightning LoRA weights: {e}")
     print("Continuing with base model...")
 
+# Apply the same optimizations from the first version
+pipe.transformer.__class__ = QwenImageTransformer2DModel
+pipe.transformer.set_attn_processor(QwenDoubleStreamAttnProcessorFA3())
+
+# --- Ahead-of-time compilation ---
+optimize_pipeline_(pipe, image=Image.new("RGB", (1024, 1024)), prompt="prompt")
+
 # --- UI Constants and Helpers ---
 MAX_SEED = np.iinfo(np.int32).max