Create app.py

app.py

@@ -0,0 +1,67 @@
+import gradio as gr
+import numpy as np
+from PIL import Image
+import cv2
+import os
+
+# Minimal Real-ESRGAN setup (CPU). This uses cv2 dnn_superres as a fallback
+# if Real-ESRGAN import fails on CPU-only. For best quality, try realesrgan.
+try:
+    from realesrgan import RealESRGAN
+    HAVE_REALESRGAN = True
+except Exception:
+    HAVE_REALESRGAN = False
+
+# Helper: upscale using Real-ESRGAN if available, else Lanczos as CPU fallback
+def upscale_core(img: Image.Image, scale: int, model_key: str) -> Image.Image:
+    if HAVE_REALESRGAN:
+        # RealESRGAN works on CPU too (slow), but OK for a free Space
+        # Model choices
+        model_map = {
+            "pro": "RealESRGAN_x4plus",
+            "standard": "RealESRNet_x4plus",
+            "creative": "RealESRGAN_x4plus_anime_6B",
+        }
+        model_name = model_map.get(model_key, "RealESRGAN_x4plus")
+        upsampler = RealESRGAN(device="cpu", scale=4)
+        # Load model
+        upsampler.load_weights(model_name)
+        # If requested scale is 2/3/4, do single pass; if >4, chain extra resize
+        primary = min(max(scale,2), 4)
+        out = upsampler.predict(np.array(img), batch_size=1)
+        out_img = Image.fromarray(out)
+        if scale > 4:
+            factor = scale / 4.0
+            w = int(out_img.width * factor)
+            h = int(out_img.height * factor)
+            out_img = out_img.resize((w,h), Image.LANCZOS)
+        return out_img
+    else:
+        # Fallback: pure CPU Lanczos upscale (not AI, but ensures API always returns)
+        w = int(img.width * scale)
+        h = int(img.height * scale)
+        return img.resize((w, h), Image.LANCZOS)
+
+def upscale(image: np.ndarray, scale: int, model: str) -> Image.Image:
+    pil = Image.fromarray(image)
+    scale = max(2, min(10, int(scale)))
+    model = (model or "pro").lower()
+    out = upscale_core(pil, scale, model)
+    return out
+
+with gr.Blocks() as demo:
+    gr.Markdown("# Open-Source Image Upscaler API (Real-ESRGAN, CPU)")
+    with gr.Row():
+        inp = gr.Image(type="numpy", label="Upload")
+        scl = gr.Slider(2, 10, value=4, step=1, label="Scale")
+        mdl = gr.Dropdown(["pro","standard","creative"], value="pro", label="Model")
+    out = gr.Image(type="pil", label="Upscaled")
+
+    btn = gr.Button("Upscale")
+    btn.click(upscale, [inp, scl, mdl], [out])
+
+    # NOTE: Your PHP will call this endpoint:
+    # POST {SPACE_URL}/api/predict
+    # JSON: {"data": ["data:image/png;base64,...", 4, "pro"]}
+
+demo.launch()