Alpha channel overlay back to RGB

edit

I need to overlay a watermark on a video stream. I am using av for flicking through the frames, but opencv for the image processing. It's a the point where I want to hand a frame back to av that I get a problem.

Here is an annotated example of what I am trying to do, and where I am having problems

container = av.open('video.mp4')
for frame in container.decode(video=0):
    background = frame.to_ndarray(format="bgr24")
    (h, w) = background.shape[:2]
    background = np.dstack([background, np.ones((h, w), dtype="uint8") * 255])

    overlay = np.zeros((h, w, 4), dtype="uint8")
    y1 = watermark_top_margin
    y2 = watermark_top_margin + watermark_height
    x1 = w - watermark_width - watermark_right_margin
    x2 = w - watermark_right_margin

    overlay[y1:y2, x1: x2] = resized_watermark

    # blend the two images together using transparent overlays
    output = background.copy()
    cv2.addWeighted(overlay, watermark_alpha, output, 1.0, 0, output)

    # blend the two images together using transparent overlays
    output = background.copy()
    cv2.addWeighted(overlay, watermark_alpha, output, 1.0, 0, output)

    #This is the problem right here vvvvvvvvvvvvvvvvvvvvvv
    output = cv2.cvtColor(output,cv2.COLOR_RGBA2RGB)
    #Here is the problem ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

The colour conversion loses any work I have done in the alpha channel. So the frame looks exactly as it did before I overlay the image.

    new_frame = av.video.VideoFrame.from_ndarray(output, format="bgr24")

So my question is this. How can I convert the image back to RGB format to create an av frame without losing the alpha channel information? This needs to be performat as it is happening in real time.