Update app.py

app.py

@@ -4,93 +4,90 @@ import gradio as gr
 from PIL import Image, ImageDraw, ImageFont
 from huggingface_hub import InferenceClient
 
-# =============== Config ===============
-HF_TOKEN = os.getenv("HF_TOKEN")  # set this in your environment
-MODEL_ID = "timbrooks/instruct-pix2pix"
+# Read your HF token from env
+HF_TOKEN = os.getenv("HF_TOKEN")
+
+# A couple of reasonable models for image-to-image editing on HF Serverless
+MODEL_CANDIDATES = [
+    "timbrooks/instruct-pix2pix",          # instruction-based edits
+    "stabilityai/stable-diffusion-2-1"     # img2img capable baseline
+]
 
 STYLES = [
-    "Modern Minimalist",
-    "Victorian",
-    "Brutalist",
-    "Mediterranean",
-    "Gothic Cathedral",
-    "Japanese Zen",
-    "Art Deco",
-    "Futuristic",
-    "Scandinavian",
-    "Meditation Retreat"
+    "Modern Minimalist", "Victorian", "Brutalist", "Mediterranean",
+    "Gothic Cathedral", "Japanese Zen", "Art Deco", "Futuristic",
+    "Scandinavian", "Meditation Retreat"
 ]
 
-# =============== Helpers ===============
 def info_image(msg: str, w: int = 768, h: int = 512) -> Image.Image:
-    """Simple image with centered text (used for errors)."""
+    """Simple dark-blue panel with centered text (for graceful errors)."""
     img = Image.new("RGB", (w, h), "#2C3E50")
     d = ImageDraw.Draw(img)
     font = ImageFont.load_default()
-    lines = []
-    words = msg.split()
-    line = ""
-    for word in words:
+    lines, line = [], ""
+    for word in msg.split():
         test = (line + " " + word).strip()
         if d.textlength(test, font=font) < (w - 40):
             line = test
         else:
-            lines.append(line)
-            line = word
-    if line:
-        lines.append(line)
-    total_h = len(lines) * 18
-    y = (h - total_h) // 2
+            lines.append(line); line = word
+    if line: lines.append(line)
+    y = (h - 18 * len(lines)) // 2
     for L in lines:
         tw = d.textlength(L, font=font)
         d.text(((w - tw) // 2, y), L, fill="white", font=font)
         y += 18
     return img
 
+def pil_to_png_bytes(img: Image.Image) -> bytes:
+    """Convert a PIL image to PNG bytes for HF InferenceClient."""
+    if img.mode not in ("RGB", "L"):
+        img = img.convert("RGB")
+    buf = io.BytesIO()
+    img.save(buf, format="PNG")
+    return buf.getvalue()
 
 def build_prompt(style: str) -> str:
     return (
         f"Transform this house into a {style} architectural style. "
-        f"Keep the same perspective and composition. Photorealistic, high detail, realistic lighting."
+        f"Keep the same layout, perspective, and composition. "
+        f"Photorealistic, high detail, realistic lighting."
     )
 
-# =============== Core ===============
 def transform_house(input_image: Image.Image, style: str):
-    """Upload a house image + style -> transformed image from HF model."""
+    """Upload a house + style -> transformed image from HF model, with status text."""
     if input_image is None:
-        return info_image("Please upload a house photo first.")
-
+        return info_image("Please upload a house photo first."), "Please upload a house photo first."
     if not HF_TOKEN:
-        return info_image("HF_TOKEN is missing. Set it in your environment before running.")
+        return info_image("Missing HF_TOKEN"), "❌ Missing HF_TOKEN (set env var HF_TOKEN)."
 
-    try:
-        client = InferenceClient(model=MODEL_ID, token=HF_TOKEN)
-        prompt = build_prompt(style)
+    img_bytes = pil_to_png_bytes(input_image)
+    client = InferenceClient(token=HF_TOKEN)
+    prompt = build_prompt(style)
+    last_err = None
 
-        # InferenceClient.image_to_image returns bytes in most versions; handle either bytes or PIL.Image
-        result = client.image_to_image(
-            prompt=prompt,
-            image=input_image,
-            negative_prompt="blurry, low quality, distorted, text, watermark, extra buildings",
-            guidance_scale=7.0,
-            image_guidance_scale=1.5,
-            num_inference_steps=30,
-        )
+    for model_id in MODEL_CANDIDATES:
+        try:
+            # Important: pass bytes, not a PIL image
+            out_img = client.image_to_image(
+                image=img_bytes,
+                prompt=prompt,
+                negative_prompt="blurry, low quality, distorted, text, watermark, extra buildings",
+                guidance_scale=7.0,
+                num_inference_steps=30,
+                # image_guidance_scale is not universally supported by the serverless backend,
+                # so we avoid it for maximum compatibility.
+                model=model_id
+            )
+            return out_img.convert("RGB"), f"✅ Success with {model_id}"
+        except Exception as e:
+            last_err = str(e)
 
-        if isinstance(result, bytes):
-            return Image.open(io.BytesIO(result)).convert("RGB")
-        elif isinstance(result, Image.Image):
-            return result.convert("RGB")
-        else:
-            return info_image("Unexpected response from the model.")
-    except Exception as e:
-        # Graceful error panel instead of crashing the app
-        return info_image(f"Generation failed: {e}")
+    return info_image("Generation failed. See status."), f"❌ Generation failed. Last error: {last_err}"
 
-# =============== UI ===============
+# -------- UI --------
 with gr.Blocks() as demo:
     gr.Markdown("# 🏠 House → Style Remix (Hugging Face)")
-
     with gr.Row():
         with gr.Column():
             inp = gr.Image(type="pil", label="Upload a house photo")
@@ -98,9 +95,10 @@ with gr.Blocks() as demo:
             btn = gr.Button("Generate", variant="primary")
         with gr.Column():
             out = gr.Image(label="Transformed House", height=480)
+            status = gr.Textbox(label="Status", interactive=False)
 
-    btn.click(fn=transform_house, inputs=[inp, style], outputs=out)
+    btn.click(transform_house, inputs=[inp, style], outputs=[out, status])
 
 if __name__ == "__main__":
-    # Run:  HF_TOKEN=your_token python app.py
+    # Run:  HF_TOKEN=hf_xxx python app.py
     demo.launch(debug=False)