Public Post: What's pG cookin'?

Hellooo! ^_^
Doing a public post today because I want as much feedback as possible on what I'm discussing today folks.

QUICK PARAGRAPH TO GET IT OUTTA MY SYSTEM CUS I'M STILL MAD:
Remember where we were last week? A barely functioning Daz but with an update downloading? That update broke Daz completely. I couldn't load ANY of my files and I was in full scale panic mode. Another patch was released that fixed this issue a couple of days later and that's all I want to mention about this whole debacle in this post because I was in a pit of misery last week xD
Daz currently chugging along = happy pG ^_^

NOW! Let's talk about what's been on my mind recently. You might have noticed Lisa is moving in the "image" above. Yes, the image has sprung to life via everyone's favorite:

Hand crafted 60fps animation! (j/k it's AI..)
 
Before anyone gets too concerned about the showcases here; none of these will be in the game. Lisa doesn't look enough like Lisa for me to ever consider it (except the portrait / selfies). I'll detail exactly why they don't currently look like Lisa and the amount of effort it'd take to get it there throughout the post - think of these gifs as a proof of concept and NOT as future material for the game.
The gifs are low quality so I can attach them to the post itself but all of the full high res mp4s are available via this link: https://gofile.io/d/agP4Na

"AI..? But.. why? Have you turned to the dark side?"

I was always on the dark side..? I've been trying to keep up to date on the AI world for quite some time but it's a lot of work because it's constantly changing and I mean constantly. I've kept my AI workflows for image creation reasonably up to date and I've remained fairly current with it, even updating my software + installing new checkpoints a month or so ago but I had a bit of time during Daz's outages and I can't sit still for two seconds so I educated myself a little on the next step up from image2image; image2video.

A totally different beast to tackle but I love a challenge. I've created + downloaded a collection of new workflows that I've been playing with this week, testing different checkpoints, new nodes, experimenting with prompts and motion setups, etc and the results are frankly unbelievable. The sheer speed that I can create these animations is incredible. If I was to try to do this myself in Daz it'd take hours to achieve something that looks infinitely worse and we're talking minutes for the AI to crank them out - the Sophie + Lisa kiss shot took around 10 seconds to prompt, 15minutes to render then 1 minute to upscale. Now THAT'S efficiency.

"Ehh.. What does this mean for the game?"

I have no idea honestly, it's far too soon to say. The list of possibilities and applications is bewildering to say the least.

First things first.. I don't have a good grasp on how the general adult gaming community likes this kind of thing and it's not something we've spoken about here in Lisa land.
Are you a fan of AI animation? Do you hate it? Are you excited about where it's going? Are you sick of it already? I genuinely don't know. I kind of exist in a bubble and don't have an ear to the ground on all things relating to adult games.
IF people are sick of this kind of thing then there's no point in me pursuing it with the intention of using it for the game but IF there is desire to see it then I can and will do so. Obviously because you're here animation isn't make or break for many of you, I am aware of that. Don't forget that my loyalty is with you, my subscribers after all x

"Won't it take a lot of time away from creating images?"

As far as I can tell; not really?

To set things up properly, yeah it'll take some time. To learn the ins and outs of everything? Yup, that's time too. Creating separate workflows for every possible use I could have? That's more time. Collecting a variety of checkpoints that are functional and creating LoRAs for the characters in the game? That's not just time, that's a rabbit hole.. But if everything's setup then that's basically most of the work done. Creating the videos themselves is as simple as pushing a button at that point.

Well, sort of..

There are a few steps in making game ready AI videos and I'll try to briefly cover them to give you some perspective:

1: Creating the reference image
-These are simply the images that the animation is built from. I'm already creating these for the game. Any single CG past, present or future can be used and animated so this step is moot.

2: Optimized workflows
-This is a 1 stop shop for each. It took me a while screwing around with gears and knobs to get something that would both work with my hardware and create passable results but there's still a quite a bit I can improve on here. Some of this is reliant on whether step 3 is possible or not.
Once these are setup and I'm happy with it, I don't think I need to touch it again, it's basically the engine that drives the whole process. I've got 5 workflows thus far, ranging from "quite basic" to "I can barely run this" so finding the sweet spot is key here and that's only gonna happen through testing the same seed across every workflow and testing prompts with that seed. As I get more experienced I can imagine myself creating individual workflows for every common state that I'd be using but that's future talk.

3: Setting up a LoRA for virtually every character
-This one is definitely the most time consuming. It's basically the process of teaching the AI every detail of every character that I can then reference in prompts. It only needs done once but I think it'd probably take a while to do everyone. I have researched it a little and where if you're creating a LoRA for still images it's largely doable and quite straightforward albeit time consuming, making a LoRA for video is still a work in progress it seems. It requires some more advanced knowledge, knowledge that I don't have yet. It also apparently functions best when feeding the learning tool videos of the character rather than still images so I'd likely have to animate a camera and render it in Daz for every character - this sounds mind-numbingly tedious to do but if it achieves the best results then it'd be worth it.
This is the lynch pin on whether or not this whole process will work or not because without it, it's too easy for the AI to lose track of the character's features and start moulding into something / someone unrecognizable which is obviously a deal breaker.
Without a LoRA if I had a scene with a clothed Lisa and I asked it to have her undress, the workflow alone has no idea what her body looks like so it just makes it up which is a big no, no. Continuity is paramount and that seems to be the trickiest part of this whole process. So instead I need a Lisa trained LoRA so the AI understands what Lisa looks like in each of her outfits and without any at all. This goes for every character but you get my point.
LoRAs can also be trained to match movement so in theory I could train a LoRA on sexual positions and use it to create more specific movements during sex scenes. It's a rabbit hole indeed but the more options and control I can get hold of, the better.

4: Rendering the animation
-Animation rendering seems to average between 10-20 minutes per 5 seconds. If I were making a "final" version of an animation, I'd bump up the # of passes on the video and it could take up to 30 minutes. Some of the workflows that I use can be used when I'm using my PC (not Daz but mostly anything else). Other workflows completely blow my PC to smithereens and render it pretty much unusable during the process because any extras running cause the system to run out of memory. I probably won't experiment with the latter because if this is gonna work, I want to set it up as a 1 button trial and error simulator whilst I'm doing other things pertaining to the game, the only variables that need edited from shot to shot is finding the sweet spot in how much movement is required in each video output and the prompts themselves. If it takes time away from CG development then I've already lost interest.
5 seconds is short but I might be able to find a way to loop the seed. I have some ideas here but I'm not sure that they'll work, but if they do.. I could, in theory render several videos and stitch them together later in adobe premiere pro resulting in a full loop effect with the ability to choice select in game and continue the animation with no cuts and no rewinding but this part is theory alone for now, I'm uncertain if it'd work.
The worst part of this section is that while you do have a fair amount of control as to what's output at the end, there's still a huge portion of this that's effectively a dice roll.

5: Video editing / rendering
-Stitching videos together and adding some panning or something to each isn't a big job but it is where I'll be able to see if the clips blend well together so it could have me doing some back and forth between the programs in attempts to get a more cohesive final video. For example rendering the final frame of one video as the first frame of the next can produce weird differences in color and exposure so the transitions themselves might be obvious.

6: Upscaling
-Once everything's done it's upscaling as the final step which is just increasing resolution + framerate. I can't render 720p videos efficiently with any of the workflows I've been using so I've been popping them out at a much lower resolution and low framerate. This obviously won't do for the game therefore they need upscaled, my first attempts of which you can see in the links. The cool thing however about upscaling and interpolation is that by adding frames to a video it gives me the opportunity to slow-mo some shots. This could be very cool in a lot of circumstances.
Anyway I'm not sure what the best practice is for this stuff yet as it's totally new software to me but I think much like the workflows, once I set it up once it'll be a very streamlined process.

To summarise:
1 = No extra effort (already have these)
2 = Moderate startup effort but only needs done once
3 = A lot of effort but only needs done once
4 = Varies
5 = Minor effort but might reveal unforeseen problems from step 4
6 = Minor startup effort but only needs done once

My point here is that; with a strong pipeline it could be incredibly simple to create animations for the game alongside the usual image structure once everything is setup correctly. Not only that but it'd be super simple to retroactively add animations to previous scenes in the game as well.

"Doesn't this seem too good to be true?"

Yes, it does. Which is why I'm apprehensive about getting too excited just yet. It requires absurdly powerful hardware to get this stuff working and I've hit brick wall after brick wall where it comes to vram already but on the other hand I've barely even got my foot in the door with it. I have a lot to learn still. It could be dud or it could be a game changer.
A big downside is the fact that upscaling is non-negotiable and as a result the videos are lower quality than rendered CGs (smoothed skin, less detail in features, etc, etc)
Another thing worth noting is that I've obviously cherry picked which animations to show here in this post, for every good output there lies 20 unusable ones so it isn't a perfect system by any stretch, I'll need to learn what shots make for good reference images and what prompts suit well for different scenarios and the only feedback I get is largely based on rng so it'll need a lot of testing and I mean a LOT to get a better grasp on things.
Creating LoRAs might be too grand a task as well because I've really only skimmed the surface of what it actually means to create one for video and regardless of all else, whether or not I can even create truly convincing stuff consistently is yet to be seen.
Creating the LoRAs might not even solve the face / body matching as well as I need it to and it's a lot of effort for an attempt..

But if it does..

And I can easily add animations to the game without losing much development time elsewhere?

That's an insane revelation.

Comparisons:

But is it good enough? I wasn't sure how my results stacked up against some of the industry tools so I purchased a whack of credits on an AI video generation website to put it to the test. I limited myself to 1 pass on my render and 1 pass on theirs because the credits aren't cheap and if I kept rerolling mine then that's just crappy data. The results are as follows:

A lot of the big AI tools like Kling or Seedance won't let you do nude images so I just used some other ones that let me generate nsfw stuff at the top of the list. The thing that struck me most in this example is how much better the steam looks in the home grown video. My video also decided to fully animate the camera which was decided on it's own lol. The faces went proper AI mode in all 3 but I think my pc churned out the best of the bunch here. I think it's an unfair fight because my workflow is setup for nsfw stuff and theirs likely isn't but the steam is the clincher for me. I'm a details man, the little things get me excited. Even when there's naked chicks on screen ^_^

Since faces were a mess I went with the first Open Exhibitionist gate image for the next one. See how it fared:

Not a whole lot to say here. The prompt was to "pull down her top" and my version pulled it up but I'll forgive it just this once because at least it isn't the stuff of nightmares like the other two xD
I love the detail it gave the underside of Lisa's top in that render. I can really see and feel that it's a stretchy material, the elasticity is so accurate. It's just so cool man, I'm in awe of this software.
I'm sure there are better AI models for this kind of stuff out there to compare against but I just picked this site from a tierlist thing I found online and for the price I paid and the results I got, I don't think it was a good return on investment ^_^

Since nudity was off the table for the big hitters and the other models weren't really doing too hot I decided to try something open ended and sfw so I prompted an image from the first photo club event with:

"3 people relaxing in an office"

Super simple and here's the results:

Google Veo3 wins the prize here but I'm not surprised given it's nearly 10x as expensive as the others lol. This scene is just amazing, it just looks like it could be real. The movement are so natural. Kling had a nice camera pan which I didn't expect, revealing more of the room without being told to. It just created it entirely randomly. Looks dope too, nice creepy painting. Love it.
Without prompting the characters to do anything, my wee render didn't really know what to do with Damien and Tara stuck in virtual freeze frame xD

Oh, well! The last one doesn't really hurt my feelings too much because it's not a scenario I'll ever find myself in; creating a scene with no real direction but it was still interesting nonetheless.

----

I've easily spent tens of thousands of hours making this game at this point and I've spent like 20 hours on AI animation so far and the results are nuts and there's no doubt that this time next year, they'll be even crazier.
The possibilities man.. Like imagine that I'm just chilling, I'm working on writing a scene or something for a patch in RPGM.. I could just swing to my other monitor, put an AI render on of an old CG from an old scene and be generating videos for old content at no cost or time investment. It's bonkers, really.

The jokes about Lisa becoming a movie might not be as far-fetched as once thought lol
(This is the point at which everyone responds with a giant "F*** YOU" because I'm so out of touch and everyone hates AI xD)

Anyway I'm gonna play around with it some more, it's nice having something that I can generate when I'm making dinner or going to get milk or something but I'm desperately interested to know your thoughts on this guys. Whether you hate it or love it, whether you think it's a nice toy to play with or the exciting future of adult gaming; I just want to hear what you're thinking.

Sorry for the length of this post btw, God damn.. I've been writing this for hours lol 
Update post? More of an update book.. 
Bordering on abuse making you guys listen to me ramble this much ^_^

Please do let me know your thoughts and I'll give you share of my love for another week <3
Thanks so much everyone!! xo

//pG