Midjourney Just Updated Its Most Useless Feature...

This video is all about testing Midjourney's latest update to a feature no one ever asked for /describe... It works a little better now, but watch to the end so you can discover it's limits with me.
Glibatree Discord (Glibacord) - Join the Community!
/ discord
Twitter:
/ glibatree
Instagram:
/ glibatree
TikTok:
/ glibatree
My Website:
glibatree.com/
Side Project:
dodecalearn.com
Here is the link to the Midjourney server:
/ discord

[@AuthorJoeyBissonOfficial]

I appreciate how you release content on updates. Looking forward to future videos.

[@glibatree]

I appreciate that!

[@suzannecarter445]

Fun! Too bad it can't be used in place of character ref but very interesting challenge!!

[@DrNothing23]

You didn't give it a chance only using one prompt. When has any LLM worked perfectly the first time with anything? I urge you to try again but use either more or longer prompts.

[@glibatree]

This was definitely very experimental phase, and by the end of the video (when the challenges started getting harder) you'll see I ended up choosing what seemed to be to be the best prompt of the lot. Obviously this format meant I wanted to move pretty quickly and show you a lot of kinds of results - so people could set their own expectations for when this feature might be useful to them.

[@domehouse79]

Imagine all 4 prompts every time. No need to be so tight with the credits. I think you’re missing the point of how imagine works.

[@bgck2798]

Thanks - the describe never was so great, so looks like it improves at least a bit :)

[@disciple786]

You have to imagine all four prompts. That’s the best way to get the best results

[@Whatchutuwant]

No they still sucks on that, I give up

[@thewebstylist]

So glad /describe finally in MJ6! That and /sref my favs

[@PortugueseLion]

Thanks a lot for reviewing so fast! It could be a nice challenge for you to take the prompts that you used and punt the original image as an image prompt in front of it. Then add the same as --sref at the end, and also add --iw 2 (and --c 0 --s 0 and --sw 500 and --w 0 ) In my exierience you get FAR better results. Also evident flaws can be corrected in the promt text. What is the important thing about /describe is if it CAN be usefull, not necessarily if it is usefull right out of the box. I am interested in your experiences. You prabably can improve on my suggestion .......

[@glibatree]

Yeah there could definitely be some cool results with that. Do you think the mission would be to recreate an original image almost exactly with these techniques? Or maybe see if, with enough tweaking Midjourney, can finally understand what a centaur is hahaha

[@LouisGedo]

👋 I've not tried the new version........ the old version did suck

[@glibatree]

Yeah it definitely needed a fair amount of development, I am not sure the language model behind this was actually updated. I have a hunch that they literally just changed the prompt it used to translate the computer vision data into a new kind of explanation.

[@LouisGedo]

@glibatree I'm actually still not impressed with the updated DESCRIBE 😔 Even ChatGPT and Microsoft CoPilot isn't great with describing most of the images I upload....... most being AI generated images. But that's probably because I'm a real stickler for things like details in context, style, and mood.

[@ThePallidor]

Words are very low bandwidth, thus /describe will NEVER work to recreate specific people that MJ doesn't know. It takes way too many parameters to describe a face in words.