Every year I post an update to show progress on my waifu project, which uploads 3d rendered news videos by waifus.
It is now year 3. After many updates to the voice and content , the project now works with TikTok.
Ive only updated a couple of videos one of which randomly went viral, ironically , it wasnt one of the uploads from my project. Im just starting upload the generated videos now, I have an example here:
Can I have feedback please on what I should focus on improving?
PS - Ive spent all my time coding, I literally dont have bandwidth/dollars to get a more normal rigged 3d model which I can use.
PSS - I have a final 'grammar' patch to improve her speaking pauses coming out this week
PSSS - Any thoughts about her vocal animation / standing animation / banner new informations / artile overlays etc and what could be better or what Ive got right would be immensely helpful, thanks.
i don't really give a shit
Its for youtube too but I havent connected it atm
If I dont connect to these platforms - what do you think I could use the software for?
Ive still yet to make enough for a cup of coffee and Im kinda running out of ideas - it was a hobby project after all
Can it tie into a news feed, like RSS or whatever?
yes it can - but Im not sure which one I should tie it too - do you have any ideas?
Sorry if I included an image i made to help promote it Im massive autistic about these coding projects and have no idea how to market them so just wanted to share what Ive made here in this thread as I do my own memes etc but I have no idea what Im doing really
I'd like to just run in locally with custom feeds.
How much would you pay for an app that did this?
AND MAKE THE WAIFU EAT HER MANGY FOOT SKIN AND NOT WASH FOR A YEAR
make it open source
bruv just spent 3 years on it mate a dudes gotta eat first
You've been fine eating cock for three years already, might as well have cybercommies help you with your anime news bitch.
"Hey Waifu, tell me the news"
>She presents the news as a newscaster in a second screen.
This would be ideal.
I made a second screen which includes images it fetches from the articles she is covering - but its a little buggy and doesnt always show images.
I would like to show video but have no idea what videos to poll from where if my program only has an article URL to trawl through ( that only contain images normally )
>waifu = cute anime girl
/a/ has become overrun by normies like OP long time ago. No one takes waifuism seriously there anymore.
i see lots of those posts. i think /a/ just hides itself in less popular threads like we do unless they want to shitpost.
this, but I can't blame op for not knowing what a waifu is, he seems nice
Make it read AI Greentexts, turn it into a shitpost reader/narrator for websites like this one.
That’s a fantastic idea thank you!
One question - where do I get ai green texts from ? Open ai?
Ive just done this using openai lmao
This is a really cool project but the voice is sort of off putting, have you tried using a different tts programs before?
yeah Ive tried them all. I can use the 15.ai one which is super emotive but it sounds like a girl having an emotional breakdown
15.ai is definitely worse.
There is a channel on garden gnometube called baby WOGUE that uses tts. The tts in his videos are fantastic, it might be worth your time to try and get in contact with him.
Thank you I believe you recommended this last time, I contacted him, he didnt reply. However I did figure out that what he does is manually "postprocess" the audio files. My whole system is trying to be automated, so I dont spend an hour each day post processing it. I dont think he has automated software for it that could be plugged into my code so I havent thought to bother him further.
Im going to try standard, instead of neural but punch up the pitch, maybe that will improve things.
rather just finish my project and make some cash to buy dumplings with
Is the script done by you? I mean what she reads from
No its pulled from URLs , the whole thing is automated including the rendering and the animation, it took ages bro
atm i just have it reading form hackernews
It could be cool if you used narural language processing to read from news and pull a summary from an "unbiased" source. It doesn't need to be neural networks, just good 'ol NPL. It could take maybe another year depending on the complexity, time and the method you choose.
have a nice day mouthbreather. you will never fit in
Did you write it from zero or used some engine?
from zero bruv but i hook into lots of stuff, including stuff I adapted.
I dont know man ship of theseus ya know
i use official voice api - i tried my own neural net but aws was the best
Im sure it is tho?
that not a cup of coffee but I will take it as a compliment thanks
the current model already summarising articles without bias - I rolled it myself mate. I have been considering using openAI to summarise stuff though.
waifu is not shorthand for "cute anime girl"
I need newsfu porn.
Long time no see my friend
hey man , I still havent finished my project, will I ever finish this damn thing and get it funded? hope you are well anon
If anyone is interested the latest upload is available here on the new tiktok account as youtube is buggy atm.
It's cool. I wish an anime girl would present to me good regular news as well. Almost all news now is shit
But I'm not sure where you're going with tiktok
Thanks man I hope it does work for you when I eventually finish it.
I simply chose tiktok becuase youtube were being massive d*cks and have requested verification or something because I changed a logo and well they havent given me access temporarily.
I know it doesnt make sense as tiktok is like 5 second videos and this presenter girl does go on for about 5 mins. but maybe I thought its a good testing ground, maybe Ill get some more subs to my youtube which barely has 100 subds . I dont know. Weirdly one of the first videos I manually threw onto tiktok went viral and I still have no idea why so thats something to bear in mind.
I hooked up my code to the wavenet on google and decided that the wavenet on AWS Polly was sounding way better. Funnily enough im so desperate for someone to stop complaining about the voice that im now using standard and elevated prosody, hoping tmrws voice will be radically different and no complaints - Ill paste it here.
Thanks for your heads up though. I literally think Ive tried everything, but nothing in the last 3 months or so. Hopefully the google AWS models incorporate emotions soon.
Regarding her face reflecting the context of the story. Before that would have been impossible but Ive just realised that after a major code overhaul this is now possible! The plan is to do what youve said, but I think the priority is first to make my system roll out news across different topics ( as that requires no changes really ). And once thats stable, then I will refine the animation.
But yeah Im not sure if the voice is the worse aspect. Can you prioritise what I should focus on guys?
Thanks again for the feedback.
Is it looking any better? D: D: D:
tell me it is bro because I go back to my nine to five lol
You're doing good, yeah the voice is somewhat robotic but that's the way it is. You should take a look at Pony Preservation Project on /mlp/ and ask there for advice. Unironically they're the most skilled guys in TTS ever.
Thanks man Ill go ask them!
>Pony Preservation Project on /mlp/
Oh shit, that's a very good call.
It's incredible how good it sounds.
Autism really is the most powerful force in the world.
Hey, just a heads up: Google's WaveNet is the most human-sounding TTS at the moment, while still being accessible enough (it's also somewhat configurable).
If costs aren't an issue (and they're not high, like 4 dollars per a million characters), you might want to use one of their female voices.
My shilling aside, an important part I suggest you focus on, is the facial expressions of the bot.
Especially if you're trying to present it in a news-presenter form, people will be constantly looking at the face of the speaker.
I'd personally start with having her emulate human expressions when pronouncing certain phonetics and then once you've got a decent start going, I'd randomize the expressions ever so slightly, so it doesn't look robotic.
After that you could link the content of the news articles to the expressions or tone of the voice, slow-speaking and serious looking when talking about accidents or war, smiling and slightly higher pitched when talking about uplifting events.
What are you using for this? It looks like Unity, how did you spend 3 years on this?
I dont think you get it, its a dynamically rendered auto project, and the context of the ai gathered articles will be presented in a emotive context relevant way.
If you till think its crap well I dont know mate its the best I got - you got anything thats better feel free to share - but thanks for feedback
lmao ok Ill add in pauses
good to hear from you anon thanks for checking in
Sounds better in the TikTok but the vid posted here was awful
It should be quick but not too quick
>- you got anything thats better feel free to share
I don't particluarly care about the TTS stuff, my pet peeve with these waifu games and projects is just how low-res and low-quality they are. This literally looks like any other Unity shite. Cloths and hairs visibly polygonial, animation's janky shits clipping thru textures like like they're from a PS2 game.
I know I'm severly autistic about this but cmon, it's the <current year> we have the hardware to do better.
>I know I'm severly autistic about this but cmon
nah your right
hoe-boive has beautiful realtime 3d models
this guys a moron
Yea, even that's much better than OP's shit. Rendering quality's decent but the model still looks dead and lifeless, rigid and boring.
Even CM3D2 did eyes much better and there's no excuse for having that primitive cloth physics. And there's STILL clipping here and there!
You do know that they have a bigger budget than op? Stop being retarded thanks.
Money doesn't make models, time does. And OP has time.
Yeah he should totally just spend another X years learning to be a professional standard 3D modeller / rigger
i was thinking about you the other day when pondering over my own projects op
i am glad to see you're still working on it
since you're on tiktok, maybe make smaller videos instead? iirc long videos tend to not to well consistently, even if one of your videos got viral it was probably more so just because the account was new
the low poly can be a part of the look if done right, older mmd models are still charming to this day
it's all about the art style
subtitles would help if people want to mute the robo voice
i didn't mean to reply like that...
have a nice day
bot response activated
thats because reddit took over it and made it into a quirky excuse to take weird merch pics
>thats because reddit took over it and made it into a quirky excuse to take weird merch pics
Serious waifuism didn't change. It's all was just people being in love with their waifu and wanting to gush about their feelings.
What happened was that mods started to forcefully change /a/ culture to suit newgays and normalgays, by banning Naruto hating, banning/forcing generals to move to other boards and simply deleting every unironic waifu thread. People just moved to alternative chans and boards.
She needs more pauses while speaking. Right now you could exchange her with Eminem and it would be similar.
Hey I remember you from 3 years ago
How much of her news update speech is by hand or procedurally generated? If it's 100% procedural, that's pretty cool.
Any long term monetization strategies?
100% with some random manual slogans thrown in for fun.
No idea how to monetise :/. Any ideas ?!?
>how to monetise
Although I'm pretty sure you won't be able to sell this as a whole, you could offer its functions through an external API.
Allow your customers to input news sources (single articles, news feeds, entire news sites, sports sections, etc.) and give them back data in real time that can be used to animate a character rig.
Hire a few 3D animators (or employ a few full-time when it really takes off) to create different kinds of 3D character models (or even realistic models, have you seen the MetaHumans framework?) and now you're not only able to sell your service stand-alone, but also custom 3D character models.
Shit man that's cool.
If she put out regular content from aggregates and got a decent following, you could offer to have her intake data from a paid source feed as well. Paid placement basically. Of course that takes getting an audience by doing it for free(tm) but in theory you could eventually be pretty hands off since it sounds like she could eventually be autonomous
>No idea how to monetise
Twitch. bro. This is a logical evolution of vtubers.
Give her a massive bouncing girl dick that wobbles as she talks, then sell the project to some wealthy tron.
how to solve the boring monotone voice: change it so she speaks random japanese gibberish, giggles and warbles and huffs like whatever hentai shit, then just add subtitles with whatever you really want to say
>Any thoughts about her vocal animation
Not sure how but I know developers have worked on ways to make certain sounds do different animations, basically automatically animating voiceovers.
Maybe, because you have the text of the speech, you have it easier.
audiobooks get it to read audiobooks
stephen fry pitchshifted
Cute egyptian goddess waifu
>Remember what Ancient Chinese Proverb said
I like it but you need to program in some pauses in her speech bro.
i have a dog and its a dog and i like dogs and this doggy dog world is dogawful
If you're willing to sell a bit of your soul you can easily make big bux getting a company to fund this, coom bait advertising or any sort of coom bait content gains big views i.e big money
I would need a big hollower base which I don’t have though 🙁
We're getting closer to Deus Ex level cyberpunk.
>AI news reader waifus
pretending to read papers to check on what to cover,
keeping eye contact when delivering certain pieces,
perhaps a little shyness
You're onto something here. Keep up the work. You're gonna make a lot of money
I have swapped out the voice - is this an improve or is this worse? Thank you:
I liked the original but the new one probably fits her avatar a bit more
Thanks man I appreciate it.
I personally find the new voice better because the text is spoken in one continuous line instead of slightly pausing after every word like the old one did.
Perhaps lower the pitch of the new voice ever so slightly?
Ive tried that but I think it starts to sound very monotone when you lower the pitch - fantastic
didnt mean to say fantastic
>Ive only updated a couple of videos one of which randomly went viral, ironically ,
can you link to the viral posts?
This tiktok got over a million views. I literally have no idea why - it only took me a couple of minutes to throw together
lol, yeah thats weird.
too bad you have to have a tiktok account to read comments, i'd like to know what the fuck people are saying
You don't have to have an account. Just click on the player and the comments will display in full screen
Seems to just be coomer zoomers
Maybe if you want newsfu to take off, you need to abandon the tech news, and do normie news.
*anime news, moronbop news, sneaker news
sell your soul anon, you know you want to.
I can see from your tiktok posts that you are gaming the algorithm for views
might be onto something - the technews stuff is too niche maybe
automated side gig?
btw no idea how im gaming any algorithm i know next to nothing how to game a trending algo
>btw no idea how im gaming any algorithm i know next to nothing how to game a trending algo
You know what you're doing you motherbich bastard.
Anime waifu's + asian 3D tits + le_meme_face
so whats the usecase here?
Thats pretty cool anon! I will say, I would prefer a more cybernetic look to the waifu to fit her voice alot better, as a TTS voice on a human always feels fake to me.
Maybe eventually youll get it big, and people can have custom news waifus in the morning giving them the news they care about.
Thanks for the solid positive reinforcement!
you might be onto something though maybe I just need to give her like a steampunk microphone and suddenly it will feel more natural
Good news everybody
Ive added a new waifu.
Im getting a weird gravity glitch on her boobs though, would she make a good addition to the roster? The model seems to be rigged appropriately
I forgot to include the link - the latest upload with the new glitchy boob waifu is at:
just add a pettanko with short hair so you never have to worry about physics again
ThX for the info where else can I learn about the main waifu tropes I know surprisingly little about waifus
Unironically coomers in /aco/ and /h/
there's probably articles on myanimelist about basic terminology
you should just watch more shows though
she looks cute like this too
TV tropes or VNDB tag summaries
If I got caught using this service for news, I'd definitely prefer someone caught me looking at a presenter dressed like this, instead of like this.
Thanks for feedback maybe I’ll try and use her as the default
Looks great, although I'd personally ditch the 2B hair by removing the hairband
Based and wizardpilled.
Any plans on improving the news -> Waifu parser?
I’ll focus on getting news from Reddit feeds across related subreddits eg finance , sport etc so there will be a multitude of programs. Each with its own waifu.
I’m wondering if I should have waifus read a summarised article from Wikipedia but not sure which ones or how I would stop triteness coming in.
I now my social news algo is as functional as can be but the other options not so sure what to do - any ideas ?
In terms of the actual speech of the bot, I beleive adding a pause between sentences / paragraphs would help. For now, it's just a stream of hard to understand AI speak that will not make it through the zoomer's short attention span.
Adding the article / post title would help immensely, and including relevant images would also help.
Take this for example:
If you actually hope to monetize this on TikTok or Youtube Shorts, you've got to compress it down to bite sized chunks to appeal to the average zoomer. Most zoomers don't get past 5 seconds, so they need to know what they're looking at in less than 5 seconds.
This. By the time Newsfu has started presenting the news, the fish video is already finished. Cut the fluff or move it to the end.
Thank you , you guys are right, Ill cut the cruft tomorrow
I dont know what this is but Ill check it out now - thanks
Thanks thats great feedback. The artitle title she discusess actually appears on the banner at the top of the screen but maybe that isnt clear enough.
Also there are images in the overlay box from the article. ( However I just noticed a bug that in todays news it hasnt been working). Its there in the old youtube videos though if you are interested in checking it out.
I would love to include video, but that wont be possible on articles. Maybe if I am doing an episode on a wikipedia article I could pull a video from youtube - unless anyone here has any better ideas... regarding non news content.
Thanks for the positive feedback I will look into that.
Also I have a super sketchy model I want to use, you guys dont seem to think it will affect my views so I will juse upload it and show it to you all now.
Hello, I am Angelica Hewiston with the Sanford & Son lawyer firm. I am glad I could catch you while you're still here. Several anonymous posters in this thread have stated they would like to be compensated for their work. You have attempted to source free guidance in this thread without any intention of paying a consulting fee.
I will give you 24 hours to cease and desist, or pay my clients a living wage.
S&S Lawyer firm.
send me picture of your breasts
Ancient Chinese Proverb:
He who doesn't pay at start, truly pays in the end.
Another 3rd waifu! Im starting to realise I had more riggable models than I realised - what do you think about this one folks? If its too hentaish will it ruin my views?
Sorry I forgot the link:
Its not too degenerate right guise?
well tiktok took it down.
I guess it was too degenerate 🙁
I'm not surprised.
Go jerk off or come back tomorrow, because your dick is currently taking control.
She has a safe-for-work variant if you can track that down and still want to use Musashi. I would kinda recommend against it because you might piss off a.) people who like azur lane and have a complex against Kancolle and b.) people who are diehard Kancolle fans.
thats super cool thanks for the headsup - I think Ill just customise the model further. I just liked it as it had boob physics built in ( which many models dont) Im all for the realism
*ship not tank
Post the videos elsewhere. I'm not using tiktok.
mpv + yt-dlp can play it without having to visit that shithole
That's Kancolle's IJN battleship Musashi (note the necklace she wears is actually the Imperial Japanese seal), based on one of the biggest battleships ever made IRL. While Kadowaka or Tanaka don't really chase after people for misuse of their models the threat of copyright infringement is real.
I got the model from a defunt game I think they made, I then removed the tank. I think that is 100% legal as there is no trademarked as Im not using their brand likeness - like the way its legal to modify a versace bag and then resell it - ( but illegal to stick a versace logo on an item youve made). I think the trademark works in the same way here.
Kantai Collection is still very much a ongoing thing in Japan. It's got some minor popularity westside for a time but it's a cult classic at this point over in Japan. I don't know how much of your argument is going to hold up in court but its likely not going to hold up at all in the court of public opinion - not that I think it would matter too much ultimately because those guys barely chase after fan works.
Ok thanks I’ll just use the base model instead then Cheers for the heads up you made a good point about the court of public opinion