@stroughtonsmith Remarkable. It even has what looks like realistic distortion from the glasses' lenses. The only thing that’s noticeable after looking at it for a while is that his left hand would have to be reeeeeally long, given that his left forearm isn't angling up.
@finestructure@stroughtonsmith also left hinge stem/temple of the eyeglass style is different from the right one, left ear v right ear. Just few more iterations to do a symmetry on human features i guess
That famous Pixar scene where the ChatGPT guy stares in disbelief at the Siri guy failing to toggle the light switch.
…maybe a bit of a stretch, but the image looked too good to not post. These characters and their personalities feel like they could go straight into production; how could you not include AI generation in your concept art process or mood boards?
How do you access Midjourney? I'm not into the whole algorithmically created art - so I'm not familiar with the actual use - is it a website or some sort of application?
@BoscoZebra you access it via Discord and logging in at midjourney.com. You can see the newbie channels and watch other peoples' prompts and images go by which might help figure it out
One thing that's very evident now is that if you are a famous creative your voice, your likeness, your art style and more are all going to live on for public use looong after you’re gone. Donated to the gestalt. Digital ‘resurrections’ used to be the realm of high end visual effects studios, but now anybody can make Freddie Mercury sing All I Want For Christmas Is You. When AI-generated video catches up to this level of fidelity, we're in danger 😂
Finding the magic keywords and phrases for compelling AI-generated art is incredibly satisfying. Every piece is like discovering something that nobody else in the world has seen before. You can easily spend hours crafting and iterating on a prompt to get the effect you're looking for; it seems silly to maintain that there's no effort or creative skill involved
Midjourney can be pretty consistent in style between runs, which makes it easier to see it as a useful tool in creative work. I’d love to see what comes out of a game jam that really embraces AI art
@stroughtonsmith Tech like Midjourney and Unreal Engine world building is already so impressive. As the tech converges for producing movies and real-time video game experiences it is going to be pretty powerful.
Typically it takes months to create a set. For the Fifth Element a scaled model was created for the city and miniatures were made for other pieces. Some of these pieces could be built with an assist from AI and then either digitally rendered or 3D printed to quickly create miniatures to make it come alive on camera in a matter of days instead of months.
I guess we've taught the AIs that we really frickin like heavy depth of field, huh? It’s almost difficult to get Midjourney to not output something like this 😂
I scrolled back through my Midjourney history to remind myself what its generated images used to look like, about 18 months ago. The difference to today (second pic) is astonishing
As a topic, I think this stuff is fascinating. If you're not keeping up with the bleeding edge on Midjourney, you would totally miss the huge improvements that have come to image generation since you last ran something like DiffusionBee on a Mac. It has gone from 'silly little images' to what I would say are stunning works of art and demonstrations of capability. I have no idea what this tech looks like one year from now, never mind ten
To get a feel for what kind of keywords will make for great Midjourney output, I like to pick an artist name and ask for a 'landscape by <name>’. A really basic prompt — effectively ‘when I say this person's name, what comes to mind?’; makes for an easy comparison point. Oftentimes generated art looks nothing like what that artist would actually create, but still conjures distinctive reusable style elements
‘Kehinde Wiley’ makes for an amazing seed for a prompt, though the generator seems to struggle with body parts (I've had to reject more images than usual to get to these few) — perhaps because the artist's own backgrounds are so busy it might be difficult for the network to separate elements?
Midjourney still lets you generate with each of its model versions, which is kinda neat! So I ran off the same prompt — "a boxer dog” — in the six major versions. V1 was released February 2022, V6 is the new alpha hotness
A tip that's new to me; Midjourney's /describe function can take an uploaded image and spit out prompts that could guide it back to making more. And you can use this to reverse-engineer images it has generated in the past, of styles or elements you don't recognize, to figure out how to make follow-ups.
There are so many features buried in slash commands that I might never come across
I love the candles on their heads in the second image. Was that part of the prompt, or is the cultural impact of World of Warcraft on the idea of kobolds just that strong?
Midjourney prompting incidentally feels like a great way to validate your alt text descriptions. Alt text for art is incredibly difficult, and I agonize over describing some of these outputs, but if you can describe something with words and have it come back and re-create basically the right image, then you know you're communicating the right concepts to paint a picture in somebody else's head. And that feels like a neat little superpower
@MrNuclearMonster nope. No more than any person who I might ask to paint something inspired by an existing artist or work; it's not even close to a 1:1 copy, ‘plagiarism' is a huge stretch. I find that whole argument ridiculous. It's a tool — use it or don’t
@stroughtonsmith It’s definitely not a 1:1 copy yet, many of the people in the “art” are either slightly off or extremely off as you noted but it seems like it could be close enough to fool people and you’ve demonstrated it’s getting “better” at the job. I wish we didn’t live in a capitalist system that’s gonna punish living artists but it sure seems like it’s going to punish them over time as their art is fed into it and the machine improves at reproducing and plagiarizing the original art.
@stroughtonsmith Man the technology is truly interesting but I just will never ever get over how fundamentally gross it is to ask a computer to copy the work of a (living!) artist and being excited as the computer gradually does a better and better job of copying
@cabel I mean, if I wanted to copy the work of a living artist, I could right-click the original and 'save as’ 😅 If I want to generate something new, inspired by the original, that's a whole other thing
@stroughtonsmith Yes, totally! At least a Save As is an exact copy of an intentional, intended work of art, existing only thanks to the creation of a skill and style forged over an individual’s literal lifetime, a singular vision that would not exist otherwise. The second one is the gross one
@bgrinter@cabel it’ll mimic a signature, sure. The art it learned from has signatures, so it hallucinates up nonsense signatures to put on top of AI art. If it’s seen a ton of a particular signature, it might even be able to approximate that too — it’s real good at the Apple logo, for example. They’re all just features and traits with weights in the end. It’s not like it houses a database of art inside the model
@stroughtonsmith i’m so glad you do these posts. As somebody that doesn’t have one of these AI generation accounts but likes to keep tabs on it, I find this super helpful.
@stroughtonsmith InvokeAI runs well on a Mac with enough ram, and civitai.com makes it easy to find and download models. Can do pretty amazing stuff with no subscription required.
@stroughtonsmith its impressive... say... take a single frame here, could Midjourney or another AI make a short scene given additional prompts for direction?
@stroughtonsmith That’s cool! I love the painting-ish form. I think I should pay a monthly fee as a Christmas gift for myself and play a bit with it, too. Merry Christmas, Steve!
@stroughtonsmith Here is my “Vančugen” GPT take on “powerless fridge,” “screaming men,” and “food in danger.” I hope this is not what you experienced today.
@stroughtonsmith Could you share a sample of the kind of prompt you’re using for these? I played around with v6 a bit today and wasn’t nearly as successful.
@stroughtonsmith@tantramar those actors writers etc were willing participants who gave consent (and paid!). Artists feeding the algorithm have not given any consent.
@jhuenne I think several games are already using AI-generated assets, text, audio, or upscaling techniques. I think AI-assisted assets are just going to become the norm, and far from reducing costs, I think they're going to increase the expectations of output/scale from studios. More assets, content, dialogue expected for the same budget as before
Add comment