Update app.py

app.py

@@ -101,7 +101,6 @@ if torch.cuda.is_available():
 print("Using device:", device)
 
 # --- Imports for Custom Pipeline ---
-# Note: These require the local 'qwenimage' folder to be present
 from diffusers import FlowMatchEulerDiscreteScheduler
 from qwenimage.pipeline_qwenimage_edit_plus import QwenImageEditPlusPipeline
 from qwenimage.transformer_qwenimage import QwenImageTransformer2DModel
@@ -129,7 +128,7 @@ pipe.load_lora_weights("tarn59/apply_texture_qwen_image_edit_2509",
                        weight_name="apply_texture_v2_qwen_image_edit_2509.safetensors",
                        adapter_name="texture-edit")
 
-# 2. Fuse Objects (Note: Filename contains non-ascii characters, handled as string)
+# 2. Fuse Objects
 pipe.load_lora_weights("dx8152/Qwen-Image-Edit-2509-Fusion",
                        weight_name="溶图.safetensors",
                        adapter_name="fuse-objects")
@@ -140,7 +139,7 @@ pipe.load_lora_weights("Alissonerdx/BFS-Best-Face-Swap",
                        adapter_name="face-swap")
 
 
-# Attempt to set Flash Attention 3 (Requires H100 or compatible setup)
+# Attempt to set Flash Attention 3
 try:
     pipe.transformer.set_attn_processor(QwenDoubleStreamAttnProcessorFA3())
     print("Flash Attention 3 Processor set successfully.")
@@ -172,7 +171,7 @@ def update_dimensions_on_upload(image):
 
 @spaces.GPU(duration=60)
 def infer(
-    input_image,
+    input_gallery_items,
     prompt,
     lora_adapter,
     seed,
@@ -181,9 +180,24 @@ def infer(
     steps,
     progress=gr.Progress(track_tqdm=True)
 ):
-    if input_image is None:
+    """
+    Input:
+    input_gallery_items: Since type="pil", this is a List[Tuple[PIL.Image, str]] or List[PIL.Image]
+    """
+    if not input_gallery_items:
         raise gr.Error("Please upload an image to edit.")
 
+    # Extract the image from the Gallery input
+    # When type='pil', Gradio Gallery returns a list of tuples (image, caption) or just images
+    first_item = input_gallery_items[0]
+    
+    if isinstance(first_item, tuple):
+        # Format is (PIL.Image, Caption)
+        input_pil = first_item[0]
+    else:
+        # Format is PIL.Image directly
+        input_pil = first_item
+
     # Map Dropdown choices to internal Adapter names
     adapters_map = {
         "Texture Edit": "texture-edit",
@@ -205,7 +219,7 @@ def infer(
     generator = torch.Generator(device=device).manual_seed(seed)
     negative_prompt = "worst quality, low quality, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, jpeg artifacts, signature, watermark, username, blurry"
 
-    original_image = input_image.convert("RGB")
+    original_image = input_pil.convert("RGB")
     width, height = update_dimensions_on_upload(original_image)
 
     result = pipe(
@@ -222,14 +236,25 @@ def infer(
     return result, seed
 
 @spaces.GPU(duration=60)
-def infer_example(input_image, prompt, lora_adapter):
-    if input_image is None:
+def infer_example(input_gallery_items, prompt, lora_adapter):
+    # input_gallery_items will be the list structure from gr.Examples
+    if not input_gallery_items:
         return None, 0
-    input_pil = input_image.convert("RGB")
-    guidance_scale = 4.0 # Slightly higher default for better adherence
-    steps = 30
-    result, seed = infer(input_pil, prompt, lora_adapter, 0, True, guidance_scale, steps)
-    return result, seed
+        
+    # When passed from gr.Examples with type="pil" and a Gallery component, 
+    # we might need to handle file paths if cache_examples=False or PIL if processed.
+    # However, since we use infer_example as the fn, we mimic the infer logic.
+    
+    # For examples with type="pil", gradio usually converts paths to PIL.
+    return infer(
+        input_gallery_items, 
+        prompt, 
+        lora_adapter, 
+        seed=0, 
+        randomize_seed=True, 
+        guidance_scale=4.0, 
+        steps=30
+    )
 
 
 css="""
@@ -247,7 +272,15 @@ with gr.Blocks(css=css, theme=steel_blue_theme) as demo:
 
         with gr.Row(equal_height=True):
             with gr.Column():
-                input_image = gr.Gallery(label="Input Images", show_label=False, type="pil", interactive=True)
+                # Changed to Gallery to support potential multi-image flows (conceptually) and match user request
+                input_image = gr.Gallery(
+                    label="Input Images", 
+                    show_label=False, 
+                    type="pil", 
+                    interactive=True,
+                    height=290,
+                    columns=1
+                )
                 
                 prompt = gr.Text(
                     label="Edit Prompt",
@@ -272,11 +305,26 @@ with gr.Blocks(css=css, theme=steel_blue_theme) as demo:
                     guidance_scale = gr.Slider(label="Guidance Scale", minimum=1.0, maximum=10.0, step=0.1, value=4.0)
                     steps = gr.Slider(label="Inference Steps", minimum=1, maximum=50, step=1, value=30)
         
+        # FIX: Correctly formatted inputs for gr.Gallery in Examples.
+        # Each example input corresponding to the Gallery component must be a LIST of images.
         gr.Examples(
             examples=[
-                ["examples/texture_sample.jpg", "Change the material of the object to rusted metal texture.", "Texture Edit"],
-                ["examples/fusion_sample.jpg", "Fuse the product naturally into the background.", "Fuse-Objects"],
-                ["examples/face_sample.jpg", "Swap the face with a cyberpunk robot face.", "Face-Swap"],
+                # Format: [ [Image_List], Prompt, Adapter ]
+                [
+                    ["examples/texture_sample.jpg"], 
+                    "Change the material of the object to rusted metal texture.", 
+                    "Texture Edit"
+                ],
+                [
+                    ["examples/fusion_sample.jpg"], 
+                    "Fuse the product naturally into the background.", 
+                    "Fuse-Objects"
+                ],
+                [
+                    ["examples/face_sample.jpg"], 
+                    "Swap the face with a cyberpunk robot face.", 
+                    "Face-Swap"
+                ],
             ],
             inputs=[input_image, prompt, lora_adapter],
             outputs=[output_image, seed],