Surely by now you've seen AI generated unicorns with 5 legs, or snowmen with 10 eyes. Machine learning can replicate images and 3D models based on word proximity, visual styles, etc, but we are clearly still getting a lot of incoherent real-world objects in those images. Is there a good tool for automated (not human-in-the-middle) testing of coherence on generated images (open source or not) or do we need to start one?