11 Comments

Wow, that rainbow pig example is wild. Somehow it never occurred to me to use emojis in my prompts.

Expand full comment
author

Yeah emoji combos are a treat in MJ (at least with v4 and 5.1). You can try mixing more than just two as well!

Expand full comment

Great writeup!

I've been having success using Midjourney Version 5.1 with the new "Raw" setting when I'm going for photo-realism. My initial - very unscientific - tests suggest that the images seem to be slightly better than with 5.0 while not being overly stylised.

Expand full comment
author

Ah, that's a great point - I haven't considered including the "Raw" 5.1 in the write-up as I expected it to be mostly equivalent to V5. Good to hear it's a slight step up!

Expand full comment
May 5, 2023Liked by Daniel Nest

Good work in Danish! No blue horses in Ukranian, but loving the redhead with castles in her hair!

Expand full comment
May 5, 2023Liked by Daniel Nest

I think the images idea is good to broaden use by non-English speaking users. Does Midjourney support other languages besides English?

Expand full comment
author

Just tried testing and the answer is "Some of them"?

Tried "blue horse" in Danish ("blå hest") and got:

https://cdn.discordapp.com/attachments/1042019128518643742/1103963643013501000/WhyTryAI_bla_hest_d5291490-0b0e-4195-abba-86708f9103c1.png

Then tried it in Ukrainian ("блакитний кінь")

https://media.discordapp.net/attachments/1042019128518643742/1103972027527872553/WhyTryAI___4d13b6e2-4d10-425b-98a7-406bc3be4d5b.webp

So looks like the cyrillic alphabet isn't quite parsed.

Expand full comment
May 5, 2023Liked by Daniel Nest

Fantastic write-up, so easy to understand with the visual accompaniments. Cheers Daniel!

Expand full comment
author

Happy you found it useful Mark!

Expand full comment
May 4, 2023Liked by Daniel Nest

Helpful tests comparing V5 with V5.1. Seems the problem with prompt accuracy is worse if anything but surely love the upstyling.

Expand full comment
author

Yeah, I'm not entirely sure how to interpret the "text accuracy" claim (I also tried to test how well V5.1 renders text in images, and that's still very much gibberish.) But if I see more concrete examples of this, I'll be sure to add them to the write-up.

Expand full comment