common-canvas

/

CommonCanvas-XL-C

like 26

The bigger issue is that it doesn't do as well on abstract noun because the captioner is less likely to tag them and instead of pick a more descriptive description. You can try a more specific prompt like a "a man standing in a room" and it works significantly better. It would be less likely to use more general concepts like "person" or other words for which a more descriptive noun exists: