Update app.py

app.py

@@ -1,14 +1,20 @@
+# --------------------------------------------------------------
+# Qwen‑Image‑Edit‑2509 LoRA Demo – fixed‑aspect‑ratio version
+# --------------------------------------------------------------
 import os
-import gradio as gr
+import random
 import numpy as np
-import spaces
 import torch
-import random
-from PIL import Image
+import gradio as gr
+import spaces
+from PIL import Image, ImageOps
 from typing import Iterable
+
+# --------------------  THEME  ---------------------------------
 from gradio.themes import Soft
 from gradio.themes.utils import colors, fonts, sizes
 
+# add a custom colour
 colors.steel_blue = colors.Color(
     name="steel_blue",
     c50="#EBF3F8",
@@ -24,6 +30,7 @@ colors.steel_blue = colors.Color(
     c950="#1E3450",
 )
 
+
 class SteelBlueTheme(Soft):
     def __init__(
         self,
@@ -33,10 +40,14 @@ class SteelBlueTheme(Soft):
         neutral_hue: colors.Color | str = colors.slate,
         text_size: sizes.Size | str = sizes.text_lg,
         font: fonts.Font | str | Iterable[fonts.Font | str] = (
-            fonts.GoogleFont("Outfit"), "Arial", "sans-serif",
+            fonts.GoogleFont("Outfit"),
+            "Arial",
+            "sans-serif",
         ),
         font_mono: fonts.Font | str | Iterable[fonts.Font | str] = (
-            fonts.GoogleFont("IBM Plex Mono"), "ui-monospace", "monospace",
+            fonts.GoogleFont("IBM Plex Mono"),
+            "ui-monospace",
+            "monospace",
         ),
     ):
         super().__init__(
@@ -75,88 +86,153 @@ class SteelBlueTheme(Soft):
             block_label_background_fill="*primary_200",
         )
 
+
 steel_blue_theme = SteelBlueTheme()
 
+# --------------------------------------------------------------
+#  Device & diagnostics
+# --------------------------------------------------------------
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-
-print("CUDA_VISIBLE_DEVICES=", os.environ.get("CUDA_VISIBLE_DEVICES"))
-print("torch.__version__ =", torch.__version__)
-print("torch.version.cuda =", torch.version.cuda)
-print("cuda available:", torch.cuda.is_available())
-print("cuda device count:", torch.cuda.device_count())
+print("CUDA_VISIBLE_DEVICES =", os.getenv("CUDA_VISIBLE_DEVICES"))
+print("torch.__version__   =", torch.__version__)
+print("torch.version.cuda  =", torch.version.cuda)
+print("cuda available?    :", torch.cuda.is_available())
+print("cuda device count :", torch.cuda.device_count())
 if torch.cuda.is_available():
-    print("current device:", torch.cuda.current_device())
-    print("device name:", torch.cuda.get_device_name(torch.cuda.current_device()))
-
+    print("current device     :", torch.cuda.current_device())
+    print("device name        :", torch.cuda.get_device_name(torch.cuda.current_device()))
 print("Using device:", device)
 
+# --------------------------------------------------------------
+#  Load the Qwen‑Image‑Edit model + LoRA adapters
+# --------------------------------------------------------------
 from diffusers import FlowMatchEulerDiscreteScheduler
 from qwenimage.pipeline_qwenimage_edit_plus import QwenImageEditPlusPipeline
 from qwenimage.transformer_qwenimage import QwenImageTransformer2DModel
 from qwenimage.qwen_fa3_processor import QwenDoubleStreamAttnProcessorFA3
 
 dtype = torch.bfloat16
-device = "cuda" if torch.cuda.is_available() else "cpu"
 
 pipe = QwenImageEditPlusPipeline.from_pretrained(
     "Qwen/Qwen-Image-Edit-2509",
     transformer=QwenImageTransformer2DModel.from_pretrained(
-        "linoyts/Qwen-Image-Edit-Rapid-AIO", # [transformer weights extracted from: Phr00t/Qwen-Image-Edit-Rapid-AIO]
-        subfolder='transformer',
+        "linoyts/Qwen-Image-Edit-Rapid-AIO",
+        subfolder="transformer",
         torch_dtype=dtype,
-        device_map='cuda'
+        device_map="cuda",
     ),
-    torch_dtype=dtype
+    torch_dtype=dtype,
 ).to(device)
 
-pipe.load_lora_weights("autoweeb/Qwen-Image-Edit-2509-Photo-to-Anime",
-                       weight_name="Qwen-Image-Edit-2509-Photo-to-Anime_000001000.safetensors",
-                       adapter_name="anime")
-pipe.load_lora_weights("dx8152/Qwen-Edit-2509-Multiple-angles",
-                       weight_name="镜头转换.safetensors",
-                       adapter_name="multiple-angles")
-pipe.load_lora_weights("dx8152/Qwen-Image-Edit-2509-Light_restoration",
-                       weight_name="移除光影.safetensors",
-                       adapter_name="light-restoration")
-pipe.load_lora_weights("dx8152/Qwen-Image-Edit-2509-Relight",
-                       weight_name="Qwen-Edit-Relight.safetensors",
-                       adapter_name="relight")
-pipe.load_lora_weights("dx8152/Qwen-Edit-2509-Multi-Angle-Lighting",
-                       weight_name="多角度灯光-251116.safetensors",
-                       adapter_name="multi-angle-lighting")
-pipe.load_lora_weights("tlennon-ie/qwen-edit-skin",
-                       weight_name="qwen-edit-skin_1.1_000002750.safetensors",
-                       adapter_name="edit-skin")
-pipe.load_lora_weights("lovis93/next-scene-qwen-image-lora-2509",
-                       weight_name="next-scene_lora-v2-3000.safetensors",
-                       adapter_name="next-scene")
-pipe.load_lora_weights("vafipas663/Qwen-Edit-2509-Upscale-LoRA",
-                       weight_name="qwen-edit-enhance_64-v3_000001000.safetensors",
-                       adapter_name="upscale-image")
+# ---- LoRA adapters -------------------------------------------------
+pipe.load_lora_weights(
+    "autoweeb/Qwen-Image-Edit-2509-Photo-to-Anime",
+    weight_name="Qwen-Image-Edit-2509-Photo-to-Anime_000001000.safetensors",
+    adapter_name="anime",
+)
+pipe.load_lora_weights(
+    "dx8152/Qwen-Edit-2509-Multiple-angles",
+    weight_name="镜头转换.safetensors",
+    adapter_name="multiple-angles",
+)
+pipe.load_lora_weights(
+    "dx8152/Qwen-Image-Edit-2509-Light_restoration",
+    weight_name="移除光影.safetensors",
+    adapter_name="light-restoration",
+)
+pipe.load_lora_weights(
+    "dx8152/Qwen-Image-Edit-2509-Relight",
+    weight_name="Qwen-Edit-Relight.safetensors",
+    adapter_name="relight",
+)
+pipe.load_lora_weights(
+    "dx8152/Qwen-Edit-2509-Multi-Angle-Lighting",
+    weight_name="多角度灯光-251116.safetensors",
+    adapter_name="multi-angle-lighting",
+)
+pipe.load_lora_weights(
+    "tlennon-ie/qwen-edit-skin",
+    weight_name="qwen-edit-skin_1.1_000002750.safetensors",
+    adapter_name="edit-skin",
+)
+pipe.load_lora_weights(
+    "lovis93/next-scene-qwen-image-lora-2509",
+    weight_name="next-scene_lora-v2-3000.safetensors",
+    adapter_name="next-scene",
+)
+pipe.load_lora_weights(
+    "vafipas663/Qwen-Edit-2509-Upscale-LoRA",
+    weight_name="qwen-edit-enhance_64-v3_000001000.safetensors",
+    adapter_name="upscale-image",
+)
 
 pipe.transformer.set_attn_processor(QwenDoubleStreamAttnProcessorFA3())
-MAX_SEED = np.iinfo(np.int32).max
 
-def update_dimensions_on_upload(image):
-    if image is None:
-        return 1024, 1024
-    
-    original_width, original_height = image.size
-    
-    if original_width > original_height:
-        new_width = 1024
-        aspect_ratio = original_height / original_width
-        new_height = int(new_width * aspect_ratio)
-    else:
-        new_height = 1024
-        aspect_ratio = original_width / original_height
-        new_width = int(new_height * aspect_ratio)
-        
-    # Ensure dimensions are multiples of 8
-    new_width = (new_width // 8) * 8
-    new_height = (new_height // 8) * 8
-    
-    return new_width, new_height
+# --------------------------------------------------------------
+#  Helper – keep aspect ratio, pad to a size accepted by the model
+# --------------------------------------------------------------
+MAX_SIDE = 1024                     # longest side we allow (model limit)
+DIVISIBLE_BY = 8                    # all dimensions must be a multiple of 8
+
+def _make_multiple(x: int, base: int = DIVISIBLE_BY) -> int:
+    """Round *down* to the nearest multiple of `base`."""
+    return (x // base) * base
+
+def prepare_image_for_pipe(pil_img: Image.Image):
+    """
+    1.  Resize the longer side to ``MAX_SIDE`` while preserving aspect‑ratio.
+    2.  Pad the resized image (using black pixels) so both dimensions are
+        multiples of 8 – exactly what the diffusion model expects.
+    3.  Return the padded image *and* the box that later lets us crop the
+        generation back to the original aspect‑ratio.
+    """
+    # ---- 1️⃣  Resize keeping aspect ratio ---------------------------------
+    w, h = pil_img.size
+    if max(w, h) > MAX_SIDE:
+        if w >= h:                     # wide image
+            new_w = MAX_SIDE
+            new_h = int(h * MAX_SIDE / w)
+        else:                          # tall image
+            new_h = MAX_SIDE
+            new_w = int(w * MAX_SIDE / h)
+    else:                              # already small enough
+        new_w, new_h = w, h
+
+    resized = pil_img.resize((new_w, new_h), Image.LANCZOS)
+
+    # ---- 2️⃣  Pad to multiples of 8 ----------------------------------------
+    pad_w = _make_multiple(new_w)
+    pad_h = _make_multiple(new_h)
+
+    # ``ImageOps.pad`` does the padding *and* keeps the image centred.
+    padded = ImageOps.pad(resized, (pad_w, pad_h), method=Image.LANCZOS, color=(0, 0, 0))
+
+    # The region of the original (resized) picture inside the padded canvas.
+    left = (pad_w - new_w) // 2
+    top = (pad_h - new_h) // 2
+    crop_box = (left, top, left + new_w, top + new_h)
+
+    return padded, crop_box, (new_w, new_h)   # padded img, where to cut, size after resize
+
+
+def crop_back_to_original(gen_img: Image.Image, crop_box, final_size):
+    """
+    ``gen_img`` is the image produced by the pipeline (same size as the padded
+    input).  We first crop to the region that corresponds to the *resized*
+    original picture and then upscale/downscale back to the exact dimensions
+    the user uploaded.
+    """
+    # 1️⃣  Crop to the region that contains the content
+    cropped = gen_img.crop(crop_box)
+
+    # 2��⃣  Resize back to the *original* dimensions (the size the user gave us)
+    return cropped.resize(final_size, Image.LANCZOS)
+
+
+# --------------------------------------------------------------
+#  Inference function (GPU‑bound)
+# --------------------------------------------------------------
+MAX_SEED = np.iinfo(np.int32).max
 
 @spaces.GPU(duration=30)
 def infer(
@@ -167,131 +243,197 @@ def infer(
     randomize_seed,
     guidance_scale,
     steps,
-    progress=gr.Progress(track_tqdm=True)
+    progress=gr.Progress(track_tqdm=True),
 ):
+    """Run a single edit – returns the edited image *with the same aspect‑ratio*."""
     if input_image is None:
         raise gr.Error("Please upload an image to edit.")
 
-    if lora_adapter == "Photo-to-Anime":
-        pipe.set_adapters(["anime"], adapter_weights=[1.0])
-    elif lora_adapter == "Multiple-Angles":
-        pipe.set_adapters(["multiple-angles"], adapter_weights=[1.0])
-    elif lora_adapter == "Light-Restoration":
-        pipe.set_adapters(["light-restoration"], adapter_weights=[1.0])
-    elif lora_adapter == "Relight":
-        pipe.set_adapters(["relight"], adapter_weights=[1.0])
-    elif lora_adapter == "Multi-Angle-Lighting":
-        pipe.set_adapters(["multi-angle-lighting"], adapter_weights=[1.0])
-    elif lora_adapter == "Edit-Skin":
-        pipe.set_adapters(["edit-skin"], adapter_weights=[1.0])
-    elif lora_adapter == "Next-Scene":
-        pipe.set_adapters(["next-scene"], adapter_weights=[1.0])
-    elif lora_adapter == "Upscale-Image":
-        pipe.set_adapters(["upscale-image"], adapter_weights=[1.0])
-        
+    # ---------- LoRA selection ----------
+    adapter_map = {
+        "Photo-to-Anime":      ["anime"],
+        "Multiple-Angles":     ["multiple-angles"],
+        "Light-Restoration":   ["light-restoration"],
+        "Relight":             ["relight"],
+        "Multi-Angle-Lighting":["multi-angle-lighting"],
+        "Edit-Skin":           ["edit-skin"],
+        "Next-Scene":          ["next-scene"],
+        "Upscale-Image":       ["upscale-image"],
+    }
+    pipe.set_adapters(adapter_map.get(lora_adapter, []), adapter_weights=[1.0])
+
+    # ---------- Seed ----------
     if randomize_seed:
         seed = random.randint(0, MAX_SEED)
-
     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")
-    
-    # Use the new function to update dimensions
-    width, height = update_dimensions_on_upload(original_image)
+    # ---------- Prompt ----------
+    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"
+    )
+
+    # ---------- Image preparation ----------
+    original = input_image.convert("RGB")
+    padded, crop_box, resized_sz = prepare_image_for_pipe(original)
 
+    # ---------- Diffusion ----------
     result = pipe(
-        image=original_image,
+        image=padded,
         prompt=prompt,
         negative_prompt=negative_prompt,
-        height=height,
-        width=width,
+        height=padded.height,
+        width=padded.width,
         num_inference_steps=steps,
         generator=generator,
         true_cfg_scale=guidance_scale,
     ).images[0]
 
-    return result, seed
+    # ---------- Crop / resize back ----------
+    final = crop_back_to_original(result, crop_box, original.size)
+
+    return final, seed
+
 
+# --------------------------------------------------------------
+#  Example helper (uses the same logic but with deterministic settings)
+# --------------------------------------------------------------
 @spaces.GPU(duration=30)
 def infer_example(input_image, prompt, lora_adapter):
-    input_pil = input_image.convert("RGB")
-    guidance_scale = 1.0
-    steps = 4
-    result, seed = infer(input_pil, prompt, lora_adapter, 0, True, guidance_scale, steps)
-    return result, seed
-
-
-css="""
-#col-container {
-    margin: 0 auto;
-    max-width: 960px;
-}
+    """Runs a quick example – 4 steps, guidance 1.0, random seed."""
+    return infer(
+        input_image,
+        prompt,
+        lora_adapter,
+        seed=0,
+        randomize_seed=True,
+        guidance_scale=1.0,
+        steps=4,
+    )
+
+
+# --------------------------------------------------------------
+#  UI
+# --------------------------------------------------------------
+css = """
+#col-container {margin: 0 auto; max-width: 960px;}
 #main-title h1 {font-size: 2.1em !important;}
 """
 
 with gr.Blocks() as demo:
     with gr.Column(elem_id="col-container"):
-        gr.Markdown("# **Qwen-Image-Edit-2509-LoRAs-Fast**", elem_id="main-title")
-        gr.Markdown("Perform diverse image edits using specialized [LoRA](https://huggingface.co/models?other=base_model:adapter:Qwen/Qwen-Image-Edit-2509) adapters for the [Qwen-Image-Edit](https://huggingface.co/Qwen/Qwen-Image-Edit-2509) model.")
+        gr.Markdown("# **Qwen‑Image‑Edit‑2509 LoRAs – Fixed Aspect Ratio**",
+                    elem_id="main-title")
+        gr.Markdown(
+            "Edit images with a variety of LoRA adapters while preserving the "
+            "original aspect‑ratio (no unexpected cropping)."
+        )
 
         with gr.Row(equal_height=True):
+            # ---------- left column ----------
             with gr.Column():
-                input_image = gr.Image(label="Upload Image", type="pil", height=290)
-                
-                prompt = gr.Text(
+                input_image = gr.Image(
+                    label="Upload Image",
+                    type="pil",
+                    height=290,
+                )
+                prompt = gr.Textbox(
                     label="Edit Prompt",
-                    show_label=True,
-                    placeholder="e.g., transform into anime..",
+                    placeholder="e.g. transform into anime…",
                 )
-
                 run_button = gr.Button("Edit Image", variant="primary")
 
+            # ---------- right column ----------
             with gr.Column():
-                output_image = gr.Image(label="Output Image", interactive=False, format="png", height=353)
-                
-                with gr.Row():
-                    lora_adapter = gr.Dropdown(
-                        label="Choose Editing Style",
-                        choices=["Photo-to-Anime", "Multiple-Angles", "Light-Restoration", "Multi-Angle-Lighting", "Upscale-Image", "Relight", "Next-Scene", "Edit-Skin"],
-                        value="Photo-to-Anime"
+                output_image = gr.Image(
+                    label="Output Image",
+                    interactive=False,
+                    format="png",
+                    height=353,
+                )
+                lora_adapter = gr.Dropdown(
+                    label="Choose Editing Style",
+                    choices=[
+                        "Photo-to-Anime", "Multiple-Angles", "Light-Restoration",
+                        "Multi-Angle-Lighting", "Upscale-Image", "Relight",
+                        "Next-Scene", "Edit-Skin",
+                    ],
+                    value="Photo-to-Anime",
+                )
+                with gr.Accordion("Advanced Settings", open=False):
+                    seed = gr.Slider(
+                        label="Seed",
+                        minimum=0,
+                        maximum=MAX_SEED,
+                        step=1,
+                        value=0,
+                    )
+                    randomize_seed = gr.Checkbox(
+                        label="Randomize Seed",
+                        value=True,
+                    )
+                    guidance_scale = gr.Slider(
+                        label="Guidance Scale",
+                        minimum=1.0,
+                        maximum=10.0,
+                        step=0.1,
+                        value=1.0,
                     )
-                with gr.Accordion("Advanced Settings", open=False, visible=False):
-                    seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0)
-                    randomize_seed = gr.Checkbox(label="Randomize Seed", value=True)
-                    guidance_scale = gr.Slider(label="Guidance Scale", minimum=1.0, maximum=10.0, step=0.1, value=1.0)
-                    steps = gr.Slider(label="Inference Steps", minimum=1, maximum=50, step=1, value=4)
-        
+                    steps = gr.Slider(
+                        label="Inference Steps",
+                        minimum=1,
+                        maximum=50,
+                        step=1,
+                        value=4,
+                    )
+
+        # ---------- examples ----------
         gr.Examples(
             examples=[
                 ["examples/1.jpg", "Transform into anime.", "Photo-to-Anime"],
                 ["examples/5.jpg", "Remove shadows and relight the image using soft lighting.", "Light-Restoration"],
-                ["examples/4.jpg", "Use a subtle golden-hour filter with smooth light diffusion.", "Relight"],
+                ["examples/4.jpg", "Use a subtle golden‑hour filter with smooth light diffusion.", "Relight"],
                 ["examples/2.jpeg", "Rotate the camera 45 degrees to the left.", "Multiple-Angles"],
                 ["examples/7.jpg", "Light source from the Right Rear", "Multi-Angle-Lighting"],
                 ["examples/10.jpeg", "Upscale the image.", "Upscale-Image"],
                 ["examples/7.jpg", "Light source from the Below", "Multi-Angle-Lighting"],
-                ["examples/2.jpeg", "Switch the camera to a top-down right corner view.", "Multiple-Angles"],
+                ["examples/2.jpeg", "Switch the camera to a top‑down right corner view.", "Multiple-Angles"],
                 ["examples/9.jpg", "The camera moves slightly forward as sunlight breaks through the clouds, casting a soft glow around the character's silhouette in the mist. Realistic cinematic style, atmospheric depth.", "Next-Scene"],
                 ["examples/8.jpg", "Make the subjects skin details more prominent and natural.", "Edit-Skin"],
-                ["examples/6.jpg", "Switch the camera to a bottom-up view.", "Multiple-Angles"],
+                ["examples/6.jpg", "Switch the camera to a bottom‑up view.", "Multiple-Angles"],
                 ["examples/6.jpg", "Rotate the camera 180 degrees upside down.", "Multiple-Angles"],
                 ["examples/4.jpg", "Rotate the camera 45 degrees to the right.", "Multiple-Angles"],
-                ["examples/4.jpg", "Switch the camera to a top-down view.", "Multiple-Angles"],
-                ["examples/4.jpg", "Switch the camera to a wide-angle lens.", "Multiple-Angles"],
+                ["examples/4.jpg", "Switch the camera to a top‑down view.", "Multiple-Angles"],
+                ["examples/4.jpg", "Switch the camera to a wide‑angle lens.", "Multiple-Angles"],
             ],
             inputs=[input_image, prompt, lora_adapter],
             outputs=[output_image, seed],
             fn=infer_example,
             cache_examples=False,
-            label="Examples"
+            label="Examples",
         )
 
-    run_button.click(
-        fn=infer,
-        inputs=[input_image, prompt, lora_adapter, seed, randomize_seed, guidance_scale, steps],
-        outputs=[output_image, seed]
-    )
+        # ---------- button action ----------
+        run_button.click(
+            fn=infer,
+            inputs=[
+                input_image,
+                prompt,
+                lora_adapter,
+                seed,
+                randomize_seed,
+                guidance_scale,
+                steps,
+            ],
+            outputs=[output_image, seed],
+        )
 
 if __name__ == "__main__":
-    demo.queue(max_size=30).launch(css=css, theme=steel_blue_theme, mcp_server=True, ssr_mode=False, show_error=True)
+    demo.queue(max_size=30).launch(
+        css=css,
+        theme=steel_blue_theme,
+        mcp_server=True,
+        ssr_mode=False,
+        show_error=True,
+    )