...of all the possible formats, it outputs.. a powerpoint presentation..? What.

I don't see the word powerpoint anywhere in https://github.com/QwenLM/Qwen-Image-Layered, I only see a code snippet saving a bunch of PNGs:

  with torch.inference_mode():
      output = pipeline(**inputs)
      output_image = output.images[0]
  
  for i, image in enumerate(output_image):
      image.save(f"{i}.png")