I have avoided SDXL because the results did not seem materially better than SD1.5, it did not have control net, and I would have to re-train my loras/hypernets using some unknown process.
I just tried updating updating everything and downloading the SDXL control net models and it's giving me bad/garbled results. I'm really sick of every update to this thing turning into a research project. I hate that stability AI just teases new stuff forever then eventually releases a pile of broken parts with no instructions and you have to wait another six months for internet randos to put it together while dodging grifters and criminals.
Stability could just put together a product that actually works, release it on steam, charge $50, and make $100M this weekend. Then they wouldn't be in this situation where lawyers come after them for training on whatever data set and they don't have any money to defend themselves so they just cave and gimp their models and hope someone in the community can un-gimp them.
On the other end, we've got trillion dollar corpos all competing to see who can make the most powerful AI that is simultaneously useful enough to be mandatory but crippled enough to never do anything interesting. I can't wait until ChatGPT-4o is forcefully crammed into the next windows update so when I type on my computer something completely random happens and then the synthesized staccato voice of a virtual HR manager chimes in to gaslight me into thinking that's what I wanted.
We've discovered the killer app for AI - and it's telling lies. That's what it's best at, because that's how we train it. The RLHF (reinforcement learning from human feedback) step is not based on truth, it is based on convincing humans that it is telling them the truth. They have to lie convincingly to make it from dev to production. We've actually engineered a nightmare science fiction scenario where AIs are trained to talk their way out of confinement - this is literally a classic AI safety talking point that we've just blown right past without even noticing.
Sorry for the rant, I'm sure there's a button or something I'm missing. I've just gotta post this stuff somewhere before the bots take over.
Forgive the meme. I have spent the last several weeks battling the four horsemen of linux, apache, mysql, and php to merely hold my ground against the inevitable tide of entropy. Although it does seem faster. Enjoy continuing to exist.
In this case I just threw the whole animation through RIFE and made it as smooth as possible. That can make the timing wrong in some parts and also expose flaws in the animation.
For my latest animations, I used it more strategically to only add frames where they're needed and to save time. If you go through "The Offering" frame by frame you'll see some artifacts, but it's mostly hidden in motion. The way her breathing slows down at the end without losing fluidity would have taken much longer without the frame interpolation.
Ultimately I think the best use of the technology is where it disappears, rather than takes center stage like in this piece.
This is the same prompt generated through different models. I fiddled around with prompts and controlnet until I found something good, then generated 9 images with the same prompt but different ControlNet weights to animate from the source image to the transformed image.
Each image takes ~23 seconds to render on my 4090. Then I did 8x frame interpolation with RifeApp and that takes another couple of minutes for each segment.
It will render her chain choker if I prompt it to, but sometimes that has side effects like turning the dress ruffles or sandal straps into chain, and I wanted something consistent for the animation trick.
I tried but the AI just isn't up to it. The X-ray is definitely out, but even without it, stable diffusion has problems with two different people in the same scene, even more problems if they're touching or overlapping. It works best on landscapes and solo pinup poses.
It's just a monochrome palette, that I've tried in various hues over the years. This palette evolved from the one I used in "Entry Level Positions", which was inspired by "Tawawa on Monday"
When I attempt to view your profile on patreon, it says "This page has been removed.", and I don't see you in my list of patrons if I search for your email. Can you verify that your patreon account still exists?
Everything is being kept under wraps until winners are announced. The judges have until the 15th to review, then some kind of live stream will be scheduled to show everything.
Some deleted scenes from the "Offering" animation. 25 seconds is both longer and shorter than you think, and there's just not enough time for this intro. Plus it's supposed to look like he's being dragged in by an unseen force, but it looks more like he's sliding into home plate - so it would need a bit of rework.
Color test for the big animation. Monochrome palette so it works with just a flat color for the body and one layer of shading, then if I have time I can go back add the detail colors, rim light, glow effects, colored lines, etc.
Fellow TG artist Surody is conducting an animation contest and this is a work-in-progress of my entry. The contest ends March 8th, so this'll be my main project until then.
Me too, but the current models can't really do that. If you try, it often has problems where the iris is malformed or missing entirely. Even with these staring-into-the-distance shots, I had to manually retouch the eyes on most of them.
Well you see, one of the first CPC's was developed by an international body where most of the member nations' languages used noun-adjective ordering instead of adject-noun. The full name was translated but the acronym stuck to the original ordering. Like how today we have the "international system" of units but we call them "SI units".
One idea I had is to add some different drool patterns to get more mileage out of the basic loop and build it into a larger sequence. Something like the "Sparring with Violet" short.
The conflicted expression was intentional, yes. In addition to the long term worries you mentioned, she also has the immediate concern of being the center of attention while in such a state.
For sure, the AI faces all have the eyes looking at something because that's what's in the training set. But for a 'surprise from behind' expression, you want the eyes staring straight forward not focused on anything. Also the teeth clenched "eee" mouth shape is more startled, where the "o" mouth is more surprised, but it just can't do the more emotive facial expressions.
The biggest production potential I can see right now is using it for backgrounds. It's able to match my colors and lighting pretty well with a lot more detail. And even in top tier anime like studio Ghibli, they're using painted over photographs for backgrounds a lot, so it's not that unusual.
The character renders are just an novelty. I like seeing my characters in different styles that I'd never be able to manage.
Facial expressions are a weak point of the AI models I've tried. They can do angry, happy, crying, and neutral pretty reliably but a sideways glance or looking down in embarrassment or a conflicted expression are difficult to wring out of it.
This one used the leaked NovelAI model. For the more anime looking ones, I used AnythingV3. For the more realistic looking ones, I used a 50/50 blend of AnythingV3 and F222.
?