Prompt Design Resources for MidJourney AI

by Shane McGeehan

Blog: Post

Breaking MidJourney with 50MP
2023-09-21

Any mad scientists here? For the past few days I have been running the same prompt over and over in an attempt to break MidJourney – and I finally succeeded… well… kinda. I created a 50 megapixel image (16,320 x 3,072px) and then MidJourney said, “Nah man, you done!” 🤣

The new pan feature is really amazing, but I was curious about its limits (and learned some things in the process). I came up with a prompt to give some cool psychedelic sci-fi output, then added “–ar 1:8” in order to make a really tall canvas. Then I simply kept panning to the right and changing colors in my prompt, nothing else. The cool thing about panning is that it keeps your resolution, but adds more with the new generation. If you zoom out or anything, it downsizes the image again, so you have to only use pan.

Anyway, after a bunch of pans the grid images stopped displaying. Maybe a size limit on Discord or something, but instead of an image I was given a URL link – that didn’t work either, lol. So to even see the images I would upscale all four, then they showed. Finally, after 3 days of prompting, I hit the wall. After reaching over 16 thousand pixels it fails every single time. I’m sure the devs are aware of a file or pixel limit on their site or Discord or something, but I got it. I also found that if I compare an image early in the process to late, the early one actually has more detail in shadows, whereas the later images are underexposed. This shouldn’t happen, but I’m thinking it is from saving an image over and over again, losing quality in the process. Interesting. So technically I could composite the old with the new and get a  great image. Okay, I’ll shut up now. Back to my mad scientist AI experiments!

(click the prompt to copy it!)

/imagine prompt: blue color::1.5 photography of a fantastical city of crystal and holographic light, on the edge of a rocky cliff, led lights, foggy, volumetric lighting, bright beams of light, insanely detailed, gothic architecture, ultra-wide angle view --ar 1:8

(click to view)

More Posts

Get Shpongled

Are you Shpongled? (if so, you rock) I've been playing around with the new cref feature that allows you to refrence facial details from another image, then randomly decided to put the Shpongle Mask in my prompts... Immediately, I laughed out loud. 🤣 Anyway, here is...

Photography to Video, in RunwayML

A good artist experiments and explores, so I try to use all kinds of AI to learn how they function. I have been toying around with Runway ML lately, which is an artificial intelligence model that generates video clips from text prompts - however - you can input your...

Be Careful Out There

Here is a very different kind of blog post. A warning, or something like that. We need to talk about something really weird that’s happening on facebook. Recently there has been a huge spike in random “pages” posting AI generated images of a person and art with stupid...

Evil Shane Styles

Evil Shane... I made a custom “Evil Shane” style, and I encourage you all to make some Evil Shane’s and share them if you are up for it, prompt below... but, before you think this is some genius science, it’s not. It’s stupid. I love messing around with new tool, plus...

Blog Search

Blog Categories

Pin It on Pinterest

Share This