INNER CODE UNIT · Python

target_h

Tencent-Hunyuan/HunyuanPortrait · inference.py:58

    target_h = y2 - y1
    
    # 1. resize the generated face back to the size of the original crop box
    generated_resized = cv2.resize(generated_crop, (target_w, target_h))
    
    # 2. if no mask is provided, create a new one
    if mask is None:
        mask = create_soft_mask((target_w, target_h))
        
    # 3. get the corresponding region in the original image
    background_region = original_img[y1:y2, x1:x2].astype(np.float32) / 255.0
    foreground_region = generated_resized.astype(np.float32) / 255.0
    
    # 4. blend: Result = Gen * Mask + BG * (1 - Mask)
    blended_region = foreground_region * mask + background_region * (1 - mask)
    
    # 5. paste back to the original image
    result_img = original_img.copy()

View source record →