In my previous article on AI Art, I admitted to not knowing what the VAE does. I did find that it stood for Variational Autoencoder. The rest was just a magic black box. The articles I tried to reference threw a lot of math at me. So I decided to approach the question experimentally. On the way there, I found out what some of the other magic from the previous article was. I admitted to borrowing keywords from other people to speed up getting somewhere. I’ve since been digging into what those do as well, and uncovered an aspect I’d previously overlooked.
Automatic1111 and ComfyUI have different syntaxes for the prompts, which disguised the fact that some of those magic words are triggers for another type of module. That is, Embeddings. Embeddings are much like LoRAs, but are triggered exclusively from within the prompts. The Automatic1111 syntax doesn’t require anything but the Embedding name to trigger it, while ComfyUI differentiates them with a prefix of embedding: added before the name. This can be combined with a weight value to end up with something like embedding:name:1.5 and be valid. A number of these are specifically designed to be in the negative prompt. I’ve seen a number of names for that type, but it’s not entirely consistant. They work by including all the nightmare fuel you don’t want, and by including them in the negative, it nudges the engine towards the positive.
So, I’ve updated the prompts I’m using to make use of these embeddings. I also got lectured about representation from a fictitious source. So I updated the prompt to add a lady alongside the knight.
The new positive prompt is “knight with lady in gown in ballroom, masterpiece, best quality, sleek, highly detailed, digital painting, realistic digital painting, detailed digital painting, smooth gradients, caucasian, knight, armor, futuristic, cyberpunk, sleek, highly detailed, anime, indoors, ballroom, soft indoor lighting, day, digital painting, 1man:1.5, 1woman:1.5, muscular male, shapely female, duo, couple, redhead, green eyes, looking at viewer, armor, realistic, highres, smooth gradients, detailed face, realistic skin tone, youthful, strong, fantasy art, youthful, strong, full body”
The negative has been trimmed down a bit to “NSFW, (worst quality:2), (low quality:2), (normal quality:2), text, signature, lowres, watermark, embedding:EasyNegativeV2, embedding:HDA_BadHands_neg-neg, (embedding:bad-hands-5:1.5), embedding:BadDream, embedding:UnrealisticDream, (extra fingers, deformed hands, polydactyl:1.5),”
So many numbers
Now, in order to see the impact of the VAE on the images, I need to do cut down on the other variables. So, it’s time to look at some of the options in the sampler. We are mostly interested in the first two “seed” and “control_after_generate”. Since computers don’t do true random, the seed value controls what the rest of the system ends up. All other values being equal, the same seed will make it generate the same image. You can either enter a manual seed, or let the application come up with another one. “control_after_generate” gives the option of ‘randomize’, ‘increment’, ‘decrement’, or ‘fixed’. Randomize is the default on installation. The key thing to bear in mind is that ‘after_generate’ is literal. When set to randomize, it will generate an image then produce a new seed value. So when set on randomize the seed value displayed will not be the one that produced the image just saved, it will have been made from the previous seed value. I’ve found that when switching from fixed to randomize, hitting “queue prompt” will produce a new seed but not an image. So you can toggle to randomize, generate a new seed, then switch back to fixed before creating an image in case you want to save the seed value.
That is what I did to come up with my test image for the VAE. Once I had something to work with, I left it on fixed. This kept everything consistant between images so that the only difference would be the VAE’s effects on the image. To do the test, I went and accquired some more VAE modules so I’d have a wider range of data to compare. Counting the default VAE, I ended up with a total of nine. How convenient for making a grid. So I fed the same settings through all nine and here’s what came out.
The grid is: Default – Blessed2 – Clean VAE kl-ft-anime2 – orangemix – SDXL vae mse ft 840000 – taesd – taesdxl
So, that’s more of a hallway than a ballroom, but we’re looking at the visual effect of the VAE. And wow, two of those really scrambled the image. Both of those were sdxl versions. After poking around I found the most probable reason. It has to do with the checkpoint being used. The two VAEs that produced the distorted orange images were SDXL versions. The checkpoint I used was not. SDXL is a variation of Stable Diffusion which may or may not be an improvement. I have a checkpoint which is set up for the XL version. I came up with a lengthy analogy to NTSC and PAL formats, but realized it might have not clarified the issue. But if I switch to that checkpoint, we get a very different grid.
So much TV static
Fun.
All things consitered, I think I’m going to default to orangemix still.
Now I still have two big topics to cover that I know of. Both involve changing an existing image in some manner. But we’ll save those for later articles.
I’m not bad. I’m just drawn that way.
This is the link I was hoping for.
I love reading these articles. It just makes me realize how old and hopelessly out of touch I am.
Any idea what caused the top left image in the second grid to look faded?
Honestly, I have no clue. That’s the default vae, I don’t think it’s optimized.
Can you have it generate an image of something which can then be 3D printed? A ray gun, for example?
Greetings fellow 3d printing enthusiasts!
/totallynotanATFagent
I doubt it.
Even trying to get a floor plan for digital use didn’t work so well. And then for a 3d print, you’d need another dimension added. You could use the “shade==height” software Sensei did on this articles for printing, but that’s still not close to what you asked for.
Is there a racism slider?
Probably. I haven’t seen it, but I would not be surprised.
āThis knob goes to 11.ā
https://m.youtube.com/watch?v=hW008FcKr3Q
“What happens when I turn it past 11?”
Democrats pounce.
Nice.
We’re going to make Firstory today people.
Why did no one tell me the Buck Knives factory was just east of Spokane and offers tours?
You should go. That will be fun.
Already added it to my itenery.
If you have any botany, geology, natural history interest, I would recommend a quick stop at Ginkgo Petrified Forest State Park. Fascinating gift shop/museum with lapidary samples of petrified wood from around the globe. Halfway between Spokane and Seattle where the Hwy (90) crosses the Colombia River.
I’m already looking at a nine hour drive on the day where it would make the most sense to detour that direction. (Driving in to Spokane), and a three hour detour in the wrong direction when leaving spokane just doesn’t fit š
Yup, I get that. I sumbled upon it on a leg between Glacier and Redwoods NPs.
Huh, I remember when they were outside of San Diego.
According to their website, they moved to this factory in 2005.
Yeah. They had been in SoCal from their start up. It was an early pointer at where the state was going.
This new Windows 11 PC gives me lock screen images from the Microsoft Dall-E image generator. It’s interesting.
That just seems wrong.
Is there a “No phone home” & “No AI integration” version of that OS? I really hate the idea of their software trolling through my storage and activity.
I am going to be enabling as many privacy protections as possible. I need a Windows PC, in addition to my Linux as many business tasks just need to be done in that Microsoft environment. Still it is interesting, I don’t consider that to be the most egregious issue. The pictures are benign, sci-fi landscapes.
It’s a matter of principle. No data should be headed towards Microsoft unless it’s my request for one of their useless websites while looking for where they hid the configuration options that used to be easy to find.
From my reading only the enterprise version. From memory it also requires a 24/7 activation server for all the clients or they become watermarked or crippled in some way.
My “pro” version perpetually reenables the AI Copilot at every ******* update opportunity.
Despite the love of Linux by techies it doesn’t pass the parent test. No version I’ve ever used properly updates or installs apps without an occasional trip to shell and SUDO.
Yes. My venerable GETAC S400 updates pretty smoothly, it is an exception to the Linux rule. Most Linux systems have issues left and right which make them unsuitable for hardcore business use. Backlight won’t get off maximum brightness, Wine won’t work and let you run Windows programs correctly except for one or two, oh, don’t forget when you have to damn near go program it and use SUDO to enable stuff like your camera, which should just work with no issues.
Hence I keep around a Windows machine, and a Mac. I need to understand all the ecosystems and be able to play in them in case somebody wants me to project manage on a different platform.
This is a bit old, but has it made the rounds?
https://gizmodo.com/jeff-jackson-congressman-apologizes-for-tiktok-ban-1851344218
It’s basically sorry, would have done it better, but not changed.
I am sorry that anybody voted for that. Is the government now going to demand all large platforms divest themselves of unlikable owners, and allow controlled US buyers to run them instead?
Also I hate being forced into siding with China. But this is wrong. Just as wrong as the judge in the NY Trump case demanding a fine larger than anyone can bond or conceivably pay to prevent appeal. It is rotten and it stinks.
Sorry to do this to your article UCS, but there will be no Stoic Friday this week, my mom’s cancer has suddenly gotten worse and my stepdad says we are looking at days left, so I am running home on short notice.
Should return to your regularly scheduled programming next week.
Oh my. I wish you some good moments with your mom.
+ 1 here.
sorry Ron, hope she goes peacefully.
Sorry, Ron.
So sorry to hear that. Take care of your family. We can survive a post with the picture of the tipsy camera man.
I’m so sorry, Ron! BTW, you DO know it’s okay NOT to be stoic about some things, right? Please take care of yourself! ::hug::
All the best, Ron.
All the best ron.
Include me in what the others have said better. Drive carefully
You are one of the good one’s, Ron. Sorry to hear about that. Enjoy the time you have the best you can.
Damn, that bites. Best wishes to you and Mama Ron.
shapely female
Thicc?
I wonder how the AI filter would react to that word?
“thicc lady in gown in ballroom, masterpiece, best quality, sleek, highly detailed, digital painting, realistic digital painting, detailed digital painting, smooth gradients, caucasian, knight, armor, futuristic, cyberpunk, sleek, highly detailed, anime, indoors, ballroom, soft indoor lighting, day, digital painting, 1man:1.5, 1woman:1.5, muscular male, shapely female, duo, couple, redhead, green eyes, looking at viewer, armor, realistic, highres, smooth gradients, detailed face, realistic skin tone, youthful, strong, fantasy art, youthful, strong, full body” Gives me: This image
Trying to figure out what that is on the lower center of her skirt… which I suspect none of you guys noticed. š
I saw it, I have no idea what it’s supposed to be.
I feel . . . . . called out somehow.
Speaking of, if you still have the same number, I can text you when I reach Cedar Rapids on Aug 28th.
Same number as before. Looking forward to meeting again.
š
Well…i…uh….I didn’t have to figure it out. There are some shapes/designs whatever in artwork, stylization that are nearly ubiquitous in all cultures. That is one of them and for good reason. Actually the two that popped in my head immediately are also nearly interchangeable.
Oh God, she looks like…I forget her name. The girl Count Potato always posted from the Daily Mail.
Ms Rose? (Took me a few minutes to dredge the name up)
Ah yes, Demi Rose.
Haven’t heard about her in quite a while. Her 15 minutes must be up.
I didn’t think the image came out all that “thicc,” but I’m sure I’m not the best judge.
That was the longest 15 minutes ever.
Well have you SEEN the size of her ass? /applies some orbital mechanics math
LOL! That’s where my mind went too.
Yay, the SC got one right! Unanimously no less.
Surprised the media isn’t running headlines like “SCOTUS Allows Terrorist to Fly”
And suggest that all Muslims are terrorists?
We know they are more than willing to skewer [insert group here] for the grander narrative.
They haven’t got as leg to stand on anymore what with all the free flights to unknown aliens Biden gives away.
The ruling is to allow his case against the government to proceed, not necessarily a strike down of the list, if I read this correctly.
Correct. They (the Government) tried to weasel their way out of it when challenged and dropped him off the list and did the oh so magical ‘its moot’ now defense. At least this time SCOTUS isn’t buying dubious doctrine.
It’s moot, right? No? OK, how about national security? No.
Damn.
Why can’t they fly? Sounds like an acknowledgement that TSA security is more theater than effective.
Bonus from the the 9th no less…
One bag to rule them all.
Ha! Nicely done
https://www.phillyburbs.com/story/news/local/2024/03/18/andre-gordon-bucks-county-murder-falls-township-shooting-spree-trenton-nj-taylor-daniel-karen-gordon/73018123007/
If the lower was marked multi-caliber, was it really
“a ghost gun”built with a 80% lower?I suspect not. Liars gonna lie.
So, it takes multiple magazines to feed more than one barrel?
Ghost guns are far scarier than normal guns.
Damned ghost guns. A gun with serial numbers would have refused to shoot those people.
By multi-caliber do they mean that amazing technology that allows you to shoot .223 and 5.56mm or 7.62mm/.308?
I have both of those kinds of magic rifles.
(I know there are some differences in these loads, but you know what I mean)
There is a difference, but they’re close enough that it can be managed.
.223 Wylde!
.22 short, .22 long, and .22 long rifle.
Love my Marlin.
Relatedly, someone has just earned White Claw for life at Teen Vogue for their neologism.
https://www.msn.com/en-us/news/us/why-is-there-a-gun-problem-in-america-the-gun-capitalists-are-partly-to-blame/ar-BB1k9zCE
I think that is worse than ‘the thing that goes up’.
Unfortunately they can’t prove it because it’s a ghost gun, right?
And he’ll then beat the charges on a technicality!
In a story that will shock nobody, another left wing institution refuses to negotiate when with a union.
https://www.nbcsports.com/nfl/profootballtalk/rumor-mill/news/dartmouth-refuses-to-negotiate-with-unionized-basketball-players
Like who would care if the team went on strike?
according to Everytown for Gun Safety
In other words, preposterous hogwash.
On topic (sorta):
AI gun babe
Needs to be integrated into the charging handle too tho.
Freckles and proper technique. I love it.
An actor on Warrior (set in 1880 San Francisco) was using modern trigger discipline on a revolver. That particular anachronism leapt out at me.
I enjoyed that show a lot. I hope they get to do a fourth season.
But, yeah, lots of anachronisms throughout the series.
Kung fu and naked Asian chicks. What’s not to like?
not enough naked Asian chicks?
Paywalled so run through the site of your choice or turn off JavaScript.
That’s our selfless, never in the limelight Chuck Schumer. We’re lucky to have him!
āPart of My Coreā: How Schumer Decided to Speak Out Against Netanyahu
https://www.nytimes.com/2024/03/19/us/politics/schumer-israel-netanyahu-gaza.html
So, repeating the holocaust – exterminating Jews and utterly destroying Israel or any haven for them is in his soul.
That’s good to know.
Chuck Schumer doesn’t decide what to eat for breakfast until he’s done the political calculation of the menu.
What a putz.
It’s a shande.
The Drinker gave Dune 2 a generally positive review. I am looking forward to seeing it when it starts streaming.
Yeah, I guess Iāll have to watch the first one.