@JapanAnon@pl.anon-kenkai.com Holy shit that sounds awesome!

@JapanAnon@pl.anon-kenkai.com cc [email protected] @noyoushutthefuckupdad@shitposter.club [email protected] [email protected]

@sjw@neckbeard.xyz @JapanAnon@pl.anon-kenkai.com Hopefully with less bare ass 🍑 Or more?

@zemichi@neckbeard.xyz @sjw@neckbeard.xyz Either case is acceptable!

@sjw@neckbeard.xyz @JapanAnon@pl.anon-kenkai.com Moon coolboymew [email protected] would simp for

@JapanAnon@pl.anon-kenkai.com cvcvcv mocap bot

@JapanAnon@pl.anon-kenkai.com What's unattainable about that?

@AllbonesJones@shitposter.club I don't have anything approaching the skillset required to realize it

@JapanAnon@pl.anon-kenkai.com @AllbonesJones@shitposter.club We mustn't let our dreams remain dreams. We need to come together to actualise this! cc [email protected] @lanodan@queer.hacktivis.me

@JapanAnon@pl.anon-kenkai.com The solution to that is a solution as old as time itself: Money. Buy your dream into reality. Pay others to get each part done.

@AllbonesJones@shitposter.club By god, you're right

@sjw@neckbeard.xyz @AllbonesJones@shitposter.club @lanodan@queer.hacktivis.me [email protected] In terms of required skills to get it off the ground, I guess it'd be: 1. 3D model rig. Can be janky, adds to the Max Headroom aesthetic 2. Mocap recordings, can be janky as above 3. System to pipe markov strings into a TTS engine 4. System to animate model/trigger mocap sequences in sync with TTS output 5. System to tie all of the above together and pipe it into OBS in way that can be streamed over RTMP

@JapanAnon@pl.anon-kenkai.com @AllbonesJones@shitposter.club @lanodan@queer.hacktivis.me [email protected] No real need to go through the trouble of piping it into obs. Let one of the many already existing solutions do that. Just raw video and PCM audio to stdout. Then you can pipe it into whatever application you want and aren't solely reliant on OBS. Especially useful if you just want to run it on a headless server. Could even pipe it to ffmpeg and then to RMTP. Could even make it an init startup service.

@sjw@neckbeard.xyz @AllbonesJones@shitposter.club @lanodan@queer.hacktivis.me [email protected] I'm just not sure how one would go about turning those various systems into raw video/audio in the first place.

@JapanAnon@pl.anon-kenkai.com @AllbonesJones@shitposter.club @lanodan@queer.hacktivis.me [email protected] Me either. Luckily there are APIs for that!

@JapanAnon@pl.anon-kenkai.com @sjw@neckbeard.xyz @lanodan@queer.hacktivis.me [email protected] 1) $0 if you make it yourself. $200-400 if you hire a guy. Plenty of free tutorials out there. 2) $20 to buy a webcam and record yourself 3) $0 dollars if you search around on stack exchange and steal code. Heck, there's free IRC markov bots you could use. 4) No idea, I've never commissioned a programmer. 5) Wouldn't standard audio and video output recorded into OBS via its natural ways of recording stuff work better? You could probably just set up a virtual camera in OBS, play the random animations on the rig, pipe that into any of several free facial recognition to vtuber programs, record THAT, and play whatever random garbage the markov bot spits out.

@JapanAnon@pl.anon-kenkai.com [email protected] @sjw@neckbeard.xyz @AllbonesJones@shitposter.club I guess you could use some mixture of YUV12 + AU or FLAC, this way it would even work on 9front. Or maybe these days it's more about pipewire but that's bloat.

@sjw@neckbeard.xyz @JapanAnon@pl.anon-kenkai.com @AllbonesJones@shitposter.club @lanodan@queer.hacktivis.me [email protected] ffmpeg can do it all!

@JapanAnon@pl.anon-kenkai.com @lanodan@queer.hacktivis.me [email protected] @sjw@neckbeard.xyz Adding to that, VUP is free on steam, does 3d, and has a way to match lip movement to the characteristics of the sound being produced. Even has free models you can screw around with to figure stuff out. You could do almost all of the free part of this by grabbing a free touhou character model and feeding it random animations and sound into a virtual camera and virtual microphone.

@AllbonesJones@shitposter.club @lanodan@queer.hacktivis.me [email protected] @sjw@neckbeard.xyz Is VUP the whole package, then? Model and sounds in, data out to virtual cam/mic?

@JapanAnon@pl.anon-kenkai.com @lanodan@queer.hacktivis.me [email protected] @sjw@neckbeard.xyz It's data out to a display of an animated model. You'd need a virtual cam/mic to feed it the max headroom stuff since it normally takes real cam and mic to apply to the model

@lanodan@queer.hacktivis.me @JapanAnon@pl.anon-kenkai.com @AllbonesJones@shitposter.club [email protected] Just raw RGB and PCM audio. Keep shit simple. Let whatever you're using to transcode/stream (likely ffmpeg) handle converting it to something else. Could also be interesting to output RGBA and let ffmpeg handle doing the background.

[email protected] @JapanAnon@pl.anon-kenkai.com @sjw@neckbeard.xyz @lanodan@queer.hacktivis.me @AllbonesJones@shitposter.club you can also render the model animations to 2D in advanced., no reason to do it on the fly and you probably don't need to use green screen, most stuff supports an alpha channel

[email protected] @JapanAnon@pl.anon-kenkai.com @lanodan@queer.hacktivis.me @AllbonesJones@shitposter.club Everyone's always talking about chroma key as if the alpha channel doesn't exist.

[email protected] @JapanAnon@pl.anon-kenkai.com @lanodan@queer.hacktivis.me @AllbonesJones@shitposter.club Yes but not live video. At least not without some kind of median. Still I think depending on OBS is extremely limiting vs just raw RGBA/PCM to stdout. If you want to use it in OBS then just pipe the output into a virtual video source so it just appears as a webcam. Plus this has the added benefit of using it with any software as well as on a headless system.

[email protected] @JapanAnon@pl.anon-kenkai.com @lanodan@queer.hacktivis.me @AllbonesJones@shitposter.club So does VP8, AV1, ProRes, DNxHD/DNxHR, and FV1 off the top of my head.

@sjw@neckbeard.xyz [email protected] @AllbonesJones@shitposter.club @JapanAnon@pl.anon-kenkai.com @lanodan@queer.hacktivis.me i would go with gpt2 btw,, still nonsencial enough to fit the aesthetic but not complete shit like markov chains.

[email protected] @AllbonesJones@shitposter.club @JapanAnon@pl.anon-kenkai.com @lanodan@queer.hacktivis.me Oh yeah, Cineform, OpenEXR, and DPX. You've also got image sequence video codecs like mj2, jxr, png, and tiff that support alpha.

@bbb@kiwifarms.cc @AllbonesJones@shitposter.club @JapanAnon@pl.anon-kenkai.com @lanodan@queer.hacktivis.me [email protected] That'd probably work pretty well.

@sjw@neckbeard.xyz @bbb@kiwifarms.cc @AllbonesJones@shitposter.club @lanodan@queer.hacktivis.me [email protected] This has developed a bit past my ability to understand it, but I'm glad to see the idea is more feasible than I thought. For what it's worth, this is something I'm willing to fund the creation of, within a budget that wouldn't bankrupt me. I'd originally thought of running it through a PeerTube Live, though I wouldn't mind unleashing it on Flote Live first as an initial test. Confusing the shit out of the userbase there would be fun too.

@JapanAnon@pl.anon-kenkai.com @sjw@neckbeard.xyz @bbb@kiwifarms.cc @lanodan@queer.hacktivis.me [email protected] The internet's first all-virtual youtuber

@AllbonesJones@shitposter.club @bbb@kiwifarms.cc @lanodan@queer.hacktivis.me [email protected] @sjw@neckbeard.xyz "We aren't controlling her. We have no idea what she's going to say. We literally can't shut her up... without KILLING HER"

@JapanAnon@pl.anon-kenkai.com @bbb@kiwifarms.cc @lanodan@queer.hacktivis.me [email protected] @sjw@neckbeard.xyz if you could weight the text generation to what gets clipped by chat or whatever, and have it learn from that, you could have the world's most existentially terrifying business model as we voluntarily replace our social interaction with AI

@JapanAnon@pl.anon-kenkai.com [email protected] @sjw@neckbeard.xyz @AllbonesJones@shitposter.club Reminds me of microsoft's AI that ended up posting white supremacist stuff because of twitter

@lanodan@queer.hacktivis.me @JapanAnon@pl.anon-kenkai.com [email protected] @sjw@neckbeard.xyz yeah, that probably where it would actually go, but also sprinkle in internet memes and erotic roleplaying "Pool's closed due to undress for mommy Hitler."

@AllbonesJones@shitposter.club @lanodan@queer.hacktivis.me [email protected] @sjw@neckbeard.xyz Now imagine how utterly based such a bot would be in VTuber form, and also not shackled by a corporation

@AllbonesJones@shitposter.club @lanodan@queer.hacktivis.me [email protected] @sjw@neckbeard.xyz First dataset to feed it would be a full database of classic copypasta

@AllbonesJones@shitposter.club @lanodan@queer.hacktivis.me @JapanAnon@pl.anon-kenkai.com [email protected] @sjw@neckbeard.xyz I already like her