The prompt:
"A space scene depicting an astronaut approaching an alien artifact in outer space. The artifact is unique and intricate, covered in hieroglyph-like symbols or a maze-like pattern. Its design is seemingly symmetrical from afar but reveals asymmetrical, non-square angles upon closer inspection. The artifact is rotating quickly, about five revolutions per minute, adding a dynamic element to the scene. The astronaut is equipped with a thruster pack and is preparing to deploy a net around the artifact. The backdrop is the vastness of space, with distant stars and perhaps a faint outline of a planet."
Dalle3's images:
Midjourney's images:
I think DE3's images are AWESOME and mj's in this case aren't getting the point. Mainly because it doesn't weigh the context "a space scene" enough, and gets confused by the word hieroglyphics.
General image comparison rules for me:
I'll use best MJ at <100$/month, at most standard settings (out of the box, --ar 1:1 etc.)
Resolution
In my judgment, mj has caught up at least 10/20 of the times (I'll generate 20 images). Catching up includes these factors:
making it a space scene
a clear approach of an alien artifact
artifact appearance
artifact rotatability
astronaut thruster pack
'preparing to deploy a net'
DE3 knocks nearly every one out, every time.
If MJ satisfies other traits of the prompt I haven't mentioned, which DE doesn't, that is worth things too.
But overall I will judge either:
NO mj has CLEARLY not caught up
YES, mj has CLEARLY caught up or surpassed this
Note: I will not update the de3 images - those are the judging standard whether or not it improves from today.
🏅 Top traders
# | Name | Total profit |
---|---|---|
1 | Ṁ20 | |
2 | Ṁ7 | |
3 | Ṁ4 | |
4 | Ṁ2 |
This seems like no to me. mj's images are a lot better than before, but relative to de3, here is my general comparison:
mj nearly never succesfully shows anything that could be interpreted as preparing to deploy a net
it does not include anything like a thruster pack in most images
the artifact doesn't look fixed to something, true, but it's not nearly as rotateable as in de3
space scene, clear approach, and general appearance sometimes are fine; appearance is slightly worse since de3's often include both the appearance of symmetry when at a distance, which is revealed to not be complete up close
Here are new DE3 samples; there are no known changes to the system from OpenAI's side, but these images now contain the original user input + the revised prompt.
I like the idea of this market, but the DALL-E images you provide don't seem to have the astronaut "preparing to deploy a net." FWIW with Midjourney v6 on Dec 24, I think it's hitting all the criteria pretty well except for the "space scene" background (which DALL-E is definitely getting) and "preparing to deploy a net."
I'm also not sure how you produced your DALL-E images. Are you sure it was based on the literal prompt you give in the criteria and not how ChatGPT reworded that prompt to help DALL-E out? Here's a typical example I get with ChatGPT:
But here is what I typically get when I send it to DALL-E itself (by prepending in ChatGPT "pass this prompt word for word with no changes at all. do not change any punctuation or spelling even a little bit. copy it exactly as it is worded here:" and checking to be sure the exact prompt was sent to the internal DALL-E API):
Though the difference isn't that big based on your criteria.