I threw a tricky prompt at some AIs and only one succeeded. Kind of.

Dall-E 3, Midjourney, Stable Diffusion, Adobe Firefly - do any of them get this prompt that humans easily understand?

It might be easy for a human to understand the prompt

"a small cat casting a shadow shaped like a lion"

But it's still a difficult prompt for even the most popular and powerful image AIs.

I gave this prompt to the four leading image AIs: Dall-E 3, Midjourney, Stable Diffusion and Adobe Firefly 2. The results of this image AI comparison were mixed.

Adobe Firefly 2 is improved in many ways but this prompt is still too tricky

Stable Diffusion. Or rather Stable Confusion

Midjourney: I’ll just ignore the part about the lion

Dall-E3. Finally a decent shadow that is kind of recognisable as a lion shape

It seems the only two AIs to really capture the idea of casting a shadow were Midjourney and Dall-E 3. And of those, only Dall-E 3 managed to create a shadow that vaguely resembles a lion.

Dall-E 3's lion shadow is not perfect but you can make out the shape of a mane and head. The other AIs failed to make shadows that looked like anything distinct.

This little test measures how well the AIs grasp concepts. Dall-E 3 has been praised for its ability to comprehend prompts, and my findings confirm this.

In conclusion, it's remarkable how much progress has occurred in just one year. That I can even give AIs a prompt like this and expect reasonable results is amazing.

I'll likely revisit this test in the future to see if the other AIs can catch up to Dall-E3.

Follow me on Instagram: @_udart_