While Iāve had good luck creating Ruby without mentioning names, Seri never comes out well if I just describe him. The actorās name greatly helps the prompt.
But I donāt keep the images that look too much like the actor, plus, in the shots where he looks very similar, heās also too old so more motivation not to use those. The ones I like the most, look nothing like the actor.
Okay, hereās my phone gallery snippet (judge away). Can you tell which actor Iām using in my prompt?
Btw, I had a moment of panic yesterday. I wanted to show my son (14) a funny meme I saved and gave him my phone but then realized that if he started scrolling through my imagesā¦ uhmmmm How do I explain why I need hundreds of images of teenage boys? Itās for a book, I swear!
I wouldnāt be able to tell who most actors are when they are under 30. The first thought that comes to mind is Zach Effron, but after his face injury and resulting surgeries on top of just getting āolder male faceā, he looks more like David Hasselhoff now.
But I could not, for the life of me, get midjourney to give me any of the epic stuff other people were generating in the discord server. I just could not figure it out. Some people were giving really specific detailed prompts. Is that the trick?
Not Zach, but I can see why youād think that. I think it helps me that the actor is a brunette and Iām changing his hair color in addition to making him a lot younger than what majority of his photos are.
Donāt try to add too many things to the image. So describe one thing in detail, give AI something concrete to latch onto. Imagine Google search results when you donāt want to get a variety of images but a ton of images that are very similar.
Experiment with keywords. For example, for Seriās images above, I wanted to give him dark blond or light brown hair but AI doesnāt get that concept. It will give me either blond or brown so I had to compromise to make him blonder than what I imagined. Similarly for Lamassu images, keyword āblack pantherā kept giving me superheroes but black lioness gave me big cats. Words matter.
Once you pick a focal thing for AI to create, add style. Iāll admit shamelessly that Iām using a real artist, Josephine Wall, because I love the results it gives me. Her artwork is colorful and imaginative and AI responds to it well. As comparison, Greg Rutkowski gives me much simpler but also nice images. But you can also use genre or art style to help guide AI.
I do recommend playing with a free one like Dream first because then you can experiment with keywords until you land on good ones and then you can try those on Midjourney or the like.
When you land on something good, keep milking it. Regenerate that prompt, maybe change a word or two and if it starts coming out too abstract, get back to the good point, put the good words back.
Okay, letās run an experiment. Iām going to play with a few prompts and get back with results.
Btw, I was feeling down last night, felt a bit triggered and wanted to make myself feel better so I started generating cats. And I love all of them but here is my fave.
Actually, you know what, let me change my avatar because this is me. Iām a cat.
I think I do prefer a commission. Itās expensive, but at least you can work closely with someone, give them examples of styles/things you like, specify how characters look/act etc. And if the artist is good, youāll get what youāre expecting or even better. Rather than trying to figure out the jackpot magic keywords thatāll generate something even remotely close to whatās in your imagination. I tried Art Breeder and the user experience was so confusing for me, I gave up lol
I spent 2 hours messing with dream doing the exact thing of your process and justā¦trying it over and over and getting annoyed with ads towards the end Itās pretty cool if you specifically know what you are looking for.
There was a couple of glitchy days that I stepped away from it. I think they were testing new presets maybe. I donāt know. Itās not the most reliable one but freeā¦
OMG. Both, Shutterstock and Adobe Stock accepted my images!
I have 3 images officially up - the same ones on both.
(yes, thatās my real name)
I donāt know how long the window for selling AI generated images is going to last. This space is very unregulated at the time and no one knows what will happen in the future. So for now, I will add a few more while theyāre accepting.
Had to:
Recolor the clothing, clone an eye, add whites/hilights to the eyes, restructure nose, lips, fingertip, copy hair stretch arm and reshade, adjust one boob a hair, fix a line in the cloth that looked like aborted lettering (and messed up the look of the cloth).
Essentially, this is the same portrait, as thereās not any major edits, except for the little sparkle/stars.
How do you bend the ai to your will? I only get weird or mediocre results. Is the key not to be specific or be super specific?
I tried Art Breeder and the user experience was so confusing for me, I gave up lol
Itās pretty cool if you specifically know what you are looking for.
