Here’s a space to share anything we’ve learned about getting the best out of Midjourney.
Pro-tip: Use /relax
if you’re on a paid plan. It’s not THAT slow and you can save your fast minutes. I think you only really need fast if you want to use the max upscale command or if you really are in a hurry, I suppose. Save those minutes!
love this scale - all the fun landmarks. Definitely gets my imagination going on where to adventure first.
The island? Yeah! There’s a definite art to crafting the prompts to get something close to what you imagine that you want and I am definitely not there yet, especially not with stuff like this! Still, I love seeing what you get out of the machine!
Originally posted to Imgur: ~200 CLIP+VQGAN keywords on 4 subjects, by @kingdomakrillic - Imgur
Those tell it specific size to output image? Very cool.
I just got my link for the beta, my cousin just opened a local gamestore here and it would be really cool to get some images printed poster size to use in the shop.
You’ll definitely need to engage in much more upscaling for a print poster size.
Found this very cool tool for upscaling images (useful for Midjourney and beyond!):
Someone put together a google doc with a list of prompts.
Here’s a big list of artist studies, using a different AI but still probably very applicable.
I started using a different AI art generator called Open AI DALL-E. It has one feature I thought you might like. After it makes an image you can erase parts of the image and have the AI redo just those parts.
For Example.
My prompt: A hand painted oil painting of the king of Saxony from the year 0936 in a wooden frame…
One of the results;
I love it! But not the wierd circle above his head or the top of his staff-septer thing. So I edited those areas out and had the AI redo.
This is one of the results:
It got rid of the weird circle and cleaned up the staff, I still would want something else on the staff but I can’t get over how great it looks.
I also wanted to try taking descriptions straight from my homebrew game and feed them into the AI.
Original Description;
The Abyssal Altarpiece
Three large wooden screens exactingly carved depicting deep-sea life. Some fish of the normal sort are depicted but crustaceans, urchins, and jellyfish predominate. The aesthetic of the piece tends to produce a feeling of unease, but the skill and artistry of the creator shine through the unearthliness of the subject matter. It was recovered from the alter of a demon worshiping cult who constructed a temple under the Dome of the Ocean (the central temple in the city of Urb) in 381.
Prompt:
An alter piece made of three large wooden screens exactingly carved depicting deep-sea life. Some fish of the normal sort are depicted but crustaceans, urchins, and jellyfish predominate. The aesthetic of the piece tends to produce a feeling of unease, but the skill and artistry of the creator shine through the unearthliness of the subject matter.
Two of the results;
What I find so interesting is how everyone got AI exactly wrong. We always thought that AI would be able to easily understand literal physical things, while it would take a much more advanced AI to understand emotions. But right now we clearly have programs that ‘understand’ what kind of wood carving would make a human feel unease. But that can’t understand that I want a wood carving with depictions of crustaceans and jellyfish.
Very cool. I applied for Dall-E2 but haven’t gotten in yet. I enjoy the process of trying to figure out what to tell the AI to get it to give me back something related to what I want. Sometimes it’s super on point and other times a lot less so, I’m fascinated by both the things it gets really well and the stuff that it has a really hard time with.
I modified and plugged your prompt into Midjourney:
altar piece, three large wooden screens, exactingly carved, deep-sea life, fish, crustaceans, urchins, jellyfish, uneasy, unearthly, terror, horror, skillfully carved
Here’s what I got back:
Wow it’s clear who gets the silver medal
At this point, I’m well over 5,000 images generated with Midjourney so I hope I’ve learned something.
Sometimes regular language works well but I find drilling down to the characteristics you want has worked for me, you could add more terms like mahogany
and worn/ruined/salvaged
to darken and dirty it up a bit potentially. I iterate a lot on prompts.