Update gradio_app.py

gradio_app.py

@@ -214,6 +214,62 @@ def update_foreground_ratio(img_proc, fr):
         foreground_res,
         gr.update(value=show_mask_img(foreground_res)),
     )
+# Clean API endpoint that doesn't depend on internal state
+def api_generate(
+    input_image,
+    foreground_ratio: float = 0.85,
+    remesh_option: str = "None",
+    vertex_count: int = -1,
+    texture_size: int = 1024,
+):
+    """
+    API endpoint for generating 3D models.
+    This endpoint handles all preprocessing internally.
+    
+    Args:
+        input_image: Input image (PIL Image)
+        foreground_ratio: Foreground ratio (0.5-1.0)
+        remesh_option: "None", "Triangle", or "Quad"
+        vertex_count: Target vertex count (-1 for auto)
+        texture_size: Texture size (512-2048)
+    
+    Returns:
+        Path to generated GLB file
+    """
+    if input_image is None:
+        raise ValueError("No image provided. Please upload an image.")
+    
+    # Preprocess the image
+    if hasattr(input_image, 'mode') and input_image.mode == 'RGBA':
+        alpha_channel = np.array(input_image.getchannel("A"))
+        if alpha_channel.min() == 0:
+            # Already has transparency, just resize
+            processed_image = sf3d_utils.resize_foreground(
+                input_image, foreground_ratio, out_size=(COND_WIDTH, COND_HEIGHT)
+            )
+        else:
+            # Need to remove background first
+            rem_removed = remove_background(input_image)
+            processed_image = sf3d_utils.resize_foreground(
+                rem_removed, foreground_ratio, out_size=(COND_WIDTH, COND_HEIGHT)
+            )
+    else:
+        # Not RGBA, need to remove background
+        rem_removed = remove_background(input_image)
+        processed_image = sf3d_utils.resize_foreground(
+            rem_removed, foreground_ratio, out_size=(COND_WIDTH, COND_HEIGHT)
+        )
+    
+    # Generate 3D model
+    if torch.cuda.is_available():
+        torch.cuda.reset_peak_memory_stats()
+    
+    glb_file = run_model(processed_image, remesh_option.lower(), vertex_count, texture_size)
+    
+    if torch.cuda.is_available():
+        print("Peak Memory:", torch.cuda.max_memory_allocated() / 1024 / 1024, "MB")
+    
+    return glb_file
 with gr.Blocks() as demo:
     img_proc_state = gr.State()
     background_remove_state = gr.State()
@@ -340,4 +396,17 @@ with gr.Blocks() as demo:
             hdr_row,
         ],
     )
+    # Register clean API endpoint
+    api_interface = gr.Interface(
+        fn=api_generate,
+        inputs=[
+            gr.Image(type="pil", label="Input Image"),
+            gr.Number(value=0.85, label="Foreground Ratio"),
+            gr.Textbox(value="None", label="Remesh Option"),
+            gr.Number(value=-1, label="Vertex Count"),
+            gr.Number(value=1024, label="Texture Size"),
+        ],
+        outputs=gr.File(label="3D Model (GLB)"),
+        api_name="generate",  # This creates /api/generate endpoint
+    )
 demo.queue().launch(share=False)