Seems to do pretty poorly with spatial relationships. "An upside down house" -> ... | Hacker News

Hacker Newsnew | past | comments | ask | show | jobs | submit

		tantalor on Aug 1, 2024 \| parent \| context \| favorite \| on: Flux: Open-source text-to-image model with 12B par... Seems to do pretty poorly with spatial relationships. "An upside down house" -> regular old house "A horse sitting on a dog" -> horse and dog next to eachother "An inverted Lockheed Martin F-22 Raptor" -> yikes https://fal.media/files/koala/zgPYG6SqhD4Y3y_E9MONu.png

colkassad on Aug 1, 2024 | [–]

Indeed: https://fal.ai/models/fal-ai/flux?share=e7e98018-fd69-45c0-9...

minimaxir on Aug 1, 2024 | | [–]

It appears the model does have some "sanity" restrictions from whatever its training process is that limits some of the super weird outputs.

"A horse sitting on a dog" doesn't work but "A dog sitting on a horse" works perfectly.

bboygravity on Aug 1, 2024 | [–]

a zebra on top of an elephant worked fine for me

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact