Background
Ten years ago I made a handful of videos analyzing SC2 but drawing principles of what made BW great. They did reasonably well as there was a large captured audience on TL in those days. But while I still stand by the insights truthfully, the videos were not great production. I wasn't good at talking and playing at the same time and really they were thrown together in an effort to get my blog word count down (which were getting pretty long at the time.)

You can still find those videos in my old blogs. However, I stopped playing SC2 and reached the limit of having meaningful analysis on a game I didn't play so the design blogs and videos stopped.

Fast forward to the pandemic and my teaching job went online. Hey! All that knowledge of capturing video via OBS (I had also experimented in livestreaming a bit) was suddenly applicable. My gaming hobby helped my job. I had also since learned how to use Davinci Resolve. I think I made 70 30-45 minutes recorded lessons in 70 days. I didn't sleep well and gained a lot of weight (two years to lose it) and gained incessant heel pain that's never truly gone away, but I got through the lockdown.

Novels and Youtube Channel
Last year, I finished my third fantasy novel and started flogging it to literary agents while starting my fourth. I started thinking maybe I could restart my youtube channel, rebranding it to analyze books and movies in fantasy and sci fi (which I know something about- certainly more than SC2 game design these days.)

Initially I thought maybe I could illustrate it using simple cartoons and bought one of those digital pads so I could hand sketch on the computer. However, I quickly realized my drawing skills had degraded since high school. I needed the images to be quick to draw but stylized enough that it looked intentionally cartoony rather than simply amateurish and bad. Maybe if I had taken art rather than band, I would have learned to draw via that shape structure method and could dash off dynamic looking drawings in a hurry. However, I already have a full time job and was trying to squash in writing on the side. To make youtube videos AND learn to cartoon was a bridge too far.

Which brought me to Midjourney
I was drawn to it because of the fantastical imagery it could create, and I initially thought I would just generate random fantasy backgrounds but I quickly started trying to match what I was saying onscreen with some sort of generated illustration... or else screen cap from videos.

I will say AI Art will not replace commissioned art any time soon. It's okay at generating something unexpected. But if you have anything in particular in mind, it's very difficult to create. Maybe Midjourney 5 is different as it is supposed to be less opinionated. However, my first forays into MJ5 were disappointing as it had lost the cool fantastical aesthetic that I loved and I wasn't sure what prompts would recapture it, so I've been sticking with MJ 4.

Examples of Limitations
I could not shift objects within an image. If you know anything about negative space from photography/ film, I was hoping to have a tower in the left third and negative space/ background in the other two thirds. No such luck. MJ will always place said tower in the centre of the image.

It's very difficult to combine separate objects in the same image and I had better success creating the elements separately and bashing them together myself on paint or Davinci's Fusion (something like Photoshop.)

Example. I had settled on Of Elves and Droids as my channel name (a metonymy of the two genres I am covering: Fantasy and Sci Fi) and I wanted to create a Janus head of an Elf facing one way and a Droid the other. Unfortunately, Midjourney kept overwriting the droid aesthetic onto both.

[image loading]

The bottom right was getting close to the morphing of heads I wanted, but there was no obvious distinction between droid on one side and elf on the other.

Fine. I'll do it myself:
[image loading]

I finally just generated an elf and droid head separately, then dropped them together and put a Canucks green circle around them.

I also wanted wanted a background image of a fantasy tower on the left and a sci-fi city on the right with some fantastical background that tied them altogether... I had to abandon the idea altogether. I tried the same trick of creating three separate elements and bashing it together myself for my first video. However, it wasn't a visually interesting piece and I ended up just reusing my logo as the background for my successive videos.

I similarly discovered some hard-coded AI definitions for a 'Knight'
I really wanted to create a black knight in the style of the Teutonic Crusaders and with a single-bladed battle axe. (Mimics the style of an old Ivanoe illustrated book with King Richard as the black knight with no sigil.)
[image loading]

Well, MJ always wants to add plate armour to its knights. No pure chainmail and crusader mantle and a great helm. No matter what I do it also wants to add a sword.
[image loading]

You can use --no as a command to eliminate certain objects.
However if you --no plate armour and --no sword
Midjourney just breaks. Error does not compute:
[image loading]

I have gotten a lot of value of making little goblins because they are just so adorably cute/ ugly. The hand/ finger problem in Midjourney 4 also doesn't really matter with goblins because they are goblins! Maybe they have really goofy looking hands.
[image loading]

Wrap Up
The last video I made turned out to have three videos worth of ideas to keep the length down. (My goal is around the 10 minute mark, but most have been 11-12 minutes.) However, I need to do something that's less labour intensive. The latest video is comprised of at least 200 separate images. And I have a folder of 2000 screen shots taken from a half dozen different films. I know it took me a full 12 hours yesterday to bring the project that had finished audio and half the illustrations to completion and published. So I suspect it took me 20 hours for that one video.

That's... not sustainable. I missed two writing days because of that (and I'm kicking myself for it.) I can do it now because I'm on Spring Break. But... I probably need to go easier on matching illustration to audio. It's fun, but time consuming. Other youtube channels have the advantage of just putting themselves on screen so they don't need visuals running all the time. But at this point, I do not wish to appear on camera, so the pictures have to do the heavy lifting.

After I finish the next two videos, I'd like to do one on my problems with Plot Armour. I hate the concept of Plot Armour and I'd like to detail why.

Well, it's been an interesting learning curve. Thanks for reading. And if you do watch my video and you enjoyed it, throw a like at it. Every little bit counts

pebble444

Italy2497 Posts

April 02 2023 20:27 GMT

Man, your really an artist; my uncle worked 20 years at skywalker sound, he knows people; send me your portfolio if you like, I will forward it to him

Falling

Canada11343 Posts

April 02 2023 21:20 GMT

I'm sure they will be ecstatic over my ability to type prompts into an AI art generator?

Zambrah

United States7243 Posts

April 03 2023 00:37 GMT

AI is so misleading, theres nothing intelligent about it, its an algorithm that just mindlessly jams images together to create an amalgamation, you've basically correctly identified why its a piss poor substitute for real artists and why its not going to replace anyone any time soon for any real length of time, lol.

Glider

United States1353 Posts

April 04 2023 08:39 GMT

On April 03 2023 06:20 Falling wrote:
I'm sure they will be ecstatic over my ability to type prompts into an AI art generator?

some advanced prompt with Midjourney can get what you are looking for without you having to do your own work. (sorry I can't provide exact parameters right now, but I have seen such images in community posts - I have pro subscription + subscription to pretty much all the decent public AI services) and Yes, vast majority of people have no idea what's going on still, most population will be impressed, for now

(not people working in the industry tho, you are right) But leverage properly, u have a couple month of head start before most people realise what's going on. Absolutely no disrespect but pebble444's reply already proved my point. He doesn't understand this area well and he is an avid internet user. Imagine most of the population.

On April 03 2023 09:37 Zambrah wrote:
AI is so misleading, theres nothing intelligent about it, its an algorithm that just mindlessly jams images together to create an amalgamation, you've basically correctly identified why its a piss poor substitute for real artists and why its not going to replace anyone any time soon for any real length of time, lol.

For midjourney (and other AI image gen available to public), for now, you are correct and this is more true than not. For now it's not gonna replace anyone, but it sure is gonna replace the tools as well as improve competency people have. Zambrah is not wrong, but just a note while this type of AI is not gonna replace anyone, the person who can use it better is gonna replace someone who don't know how to use it, or use it poorly.

But for GPT, its logical reasoning ability is getting pretty intelligent. Getting top 10% on bar exam and able to calculate and answer mathematical questions from a rough chalk board and able to tell you HOW it is making such reasoning deduction, is getting into territory of a type of intelligence. (although GPT is always better at programming, only the Chat part brought it to public attention) I don't see it takes too long for those two to combine (image + reasoning, think GPT using midJourney). And yeah it's very different still from our intelligence which is create and deduce more information from less data set (complexity out of simplicity), it is still a type intelligence.

Real general intelligence, or an AI that can learn to learn... is a little too scary to think about right now -_-

But most should know GPT can now perform tasks and on its own, tricked a human to enter a captcha to continue its over all task goal, and reasoned by itself, that it is better to lie to the human it hired as a better cause of action (it told the human it has sight problems so can't see clearly), and succeeding... this is not completely unintelligent. And things are improving at exponential rate. So don't even think about image gen or text gen, we are already at text to video gen1 and action oriented AI gen1. Basically chat-GPT can only give you text results, however useful and sophisticated it can be, but there are already AI in the works that can actually DO things for you instead of give you a list of instructions. (search Adept AI, right now invitation only, and no it's not real general intelligence despite what website says, but it will replace even more human action)

Basically, I just type, open a new google spreadsheet, give it this title, and create this dataset in column and it will actually do the tasks. "goto zillow, find me a house that fits my budget of $$$ and have a swimming pool, 3 bedroom in this zip code, then bookmark it", it will open browser and do all those things.

those are already in the works, they are here, refining to near perfection is inevitable. The next step is action will be combined with GPT level logical reasoning. "make me a viral youtube video about w/e, and upload it to youtube, with thumbnail and all meta information, based on best retention/viral trends", and it will just do it.

What will happen then I wonder, and this is not years away, it's happening now and it is month away.

Alakaslam

United States17336 Posts

April 07 2023 06:22 GMT

Too good, gotta find that like and subscribe