[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip / qa] [cm / hm / lgbt / y] [3 / aco / adv / an / asp / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / qst / sci / soc / sp / tg / toy / trv / tv / vp / wsg / wsr / x] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/mlp/ - Pony

[Advertise on 4chan]


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: AltOP.png (1.54 MB, 2119x1500)
1.54 MB
1.54 MB PNG
TwAIlight welcomes you to the Pony Voice Preservation Project!
https://clyp.it/tm03e5en

This project is the first part of the "Pony Preservation Project" dealing with the voice.
It's dedicated to saving our beloved pony's voices by creating a neural network based Text To Speech for our favorite ponies.
Videos such as https://youtu.be/GuJKTodX1FA. or https://youtu.be/DWK_iYBl8cA have proven that we now have the technology to generate convincing voices using machine learning algorithms "trained" on nothing but clean audio clips.
With roughly 10 seasons (9 seasons and 5 movies) worth of voice lines available, we have more than enough material to apply this tech for our deviant needs.

Any anon is free to join, and many are already contributing. Just read the guide to learn how you can help bring on the wAIfu revolution. Whatever your technical level, you can help.
Document: https://docs.google.com/document/d/1xe1Clvdg6EFFDtIkkFwT-NPLRDPvkV4G675SUKjxVRU

We now have a working TwAIlight that any Anon can play with:
https://fifteen.ai/
https://derpy.me/vCzm2 (48KHz Training)
https://derpy.me/hdJQF (48KHz Synthesis)
https://derpy.me/NR7Xi (Ngrok Synthesis)
https://derpy.me/YTJ94 (Guide)

>Active Tasks
Researching alternative vocoders
Anon transcribing books and comics
Making AI creations (In thread)
Cookie is working on controllable speech
Research into animation AI

>Latest Developments
More AI creations
Sound effects collection
Updates to Synthbot.ai
Small corrections to Big Mac's dataset
Synthbot making progress on AI animation
15 re-releases main site (15.ai)

>Voice samples
https://derpy.me/fHs3K
https://derpy.me/O1xdh

>Clipper Anon's Master File 2.0:
https://mega.nz/#F!L952DI4Q!nibaVrvxbwgCgXMlPHVnVw
https://mega.nz/folder/0UhSmYAB#WBrB-qCprQTofkAhwMp5CQ

>Synthbot's Torrent Resources
https://derpy.me/ZJNca

>Cool, where is the discord/forum/whatever unifying place for this project!?
You're looking at it.

Last Thread:
>>36374498
>>
FAQs:
>READ THE DOC
Do it now
derpy.me/V7cMp

>Where can I find things made with the voice AI?
In the Good Poni Content folder: derpy.me/23EUs

>Did you know that such and such voiced this other thing?
Yes. We are very much aware. It is best to keep to official audio only unless there is very little of it available. If you know of a good source of audio for characters with few (or just fewer) lines, please post it in the thread. 5.1 is generally required unless you have a source already clean of background noise. Preferably post a sample or link. The easier you make it, the more likely it will be done.

>What about fan-imatitions of official voices?
No.

>How do I make the voices?
Several guides are available. In depth guides on how to do training and synthesis (making the ponies speak) are in the doc. If you don't want to use the navigation bar in the doc, the sections are also directly linked in the OP. If you want to use the WiP 48KHz notebook, some kind Anons have put together some image guides for you.
48KHz Training: derpy.me/wW2hX
48KHz Sythesis: derpy.me/j4MXQ

>How do I make the ngrok links?
Doc: derpy.me/SfIhY
Video: derpy.me/qYgIp

>Where are all the voice samples?
In the doc.

>Is a place I can find all the pony models?
In the doc.

>What about muh waifu?
Check the doc.

>Will you guys be doing a [insert language here] version of the AI?
Probably not, but you're welcome to. You can however get most of the way there by using phoenetic transcriptions of other languages.

>What about [insert OC here]'s voice?
Not a priority. Again, however, you're welcome to. There are already people doing this.

>Where can I view the PPP /mlp/con panel?
YouTube: youtu.be/WtuKBm67YkI
CyTube chat: pony.tube/videos/watch/b83fbbfc-6d4e-4768-8deb-edb61ea38abb

>I have an idea!
Great. Post it in the thread and we'll discuss it.

>Do you have a Code of Conduct?
Of course: fifteen.ai/code

>Is this project open source? Who is in charge of this?
derpy.me/CQ3Ca
>>
File: Anch.jpg (29 KB, 500x500)
29 KB
29 KB JPG
>>36390027
Anchor.
>>
just repeating my question from last thread here
>>36389426
>>36389476
>>
https://u.smutty.horse/lyztzqpzjpu.wav
>>
File: shitpost (28).png (787 KB, 2330x1960)
787 KB
787 KB PNG
first
>>
do you think 15 actually goes on here or is it just an impersonator and he doesn't care about being impersonated?
>>
>>36390193
yes
>>
>>36390193
yes
>>
>>36390193
yes
>>
>>36390206
>>36390329
>>36390334
Alright. To find out the real 15, I want each of you to say something only he'd know.
>>
>>36390193
yes
>>
>>36390338
Ask them what number comes before 15.
>>
>>36390362
Pretty sure it's 92.
>>
>>36390193
>>36390338
https://u.smutty.horse/lyzulqtonyf.wav
>>
>>36390383
ngl, that does sound like something he'd say
>>
>>36390383
>>36390391
I wonder who 15’s waifu is
I’d guess Twiggles
>>
>>36390338
I love SFM futa anthro porn.
>>
File: fluttershy.png (238 KB, 1600x1154)
238 KB
238 KB PNG
https://voca.ro/1gAbUhJnrXJk
>>
>>36390151
Yeah I have no idea what I'm working with. I'm going to need to see what the directories look like.

Can you open up a command prompt, type without quotations, "cd Desktop" press enter, then type "tree /f C:\(I don't know what folder the game is stored in so substitute the path here without parentheses) > cp77.txt" and press enter? This will dump the names and extensions of all the game files/folders in a single text file on your desktop. Open it and make sure you dumped the correct folder, then upload the .txt to smutty.

I said I'm not experienced in this so I guarantee nothing, but I can try to help. Can you ask anyone else? I assume you asked /g/ first?
>>
Speaking of 15, when is he gonna reply to my email? I even sent him transcripts and everything.
>>
>>36390783
>I even sent him transcripts and everything.
Could I have a copy too?
>>
>>36390797
they're nothing special but sure, here you go:
https://mega.nz/folder/yaZEWRiD#r6h-InXeUHdG0Mdiw4R_Gg
https://mega.nz/folder/KKYHBaiZ#F0rs_oqbf51S5Z723PIFLg
Second one is more of a joke, but I was genuinely curious if he'd be able to make it work.
>>
File: paste.png (105 KB, 1299x1965)
105 KB
105 KB PNG
>>36390804
Thanks Anon!
Black Mesa Announcer is already in 15s list [pic related] so I think you'll get your wish one-day™.
The 2nd dataset looks both too noisy and too small. I agree that it'd be a challenge to make it work but good luck(?).
>>
>>36390843
I saw that, but the black mesa announcer is already available on the site (with a very small dataset) so I thought maybe 15 got an incomplete dataset and figured I'd send him the full thing.
>>
File: Capture.png (98 KB, 856x600)
98 KB
98 KB PNG
>>36390843
>>36390849
here, this picture should help.
>>
>>36390853
Yeah, I see thanks. Your dataset only contains a single audio file that has more than 1 word spoken so that's a possible reason the entire thing wasn't used.
I don't know about 15 but my personal model fails to align if I don't filter out short audio files.
Also, 'lab.wav' and 'usa.wav' are mis-transcribed.
>>
>>36390865
>I don't know about 15 but my personal model fails to align if I don't filter out short audio files.
that's weird. one would think this word-by-word approach would be more efficient, computationally.
>Also, 'lab.wav' and 'usa.wav' are mis-transcribed.
Thanks for catching that. I mostly just went off the filenames, and it's really easy to miss a subtle error like that.
>>
>>36390193
Joking aside, 15 does actually visit this board on occasion, whether to send a link to the latest version of the site or to share his latest developments, such as what improved performance of the models vs. what made them less effective. He visited this board well over a year ago to show off the very first iteration of his models, around November 2019.

He last visited the board yesterday I think to announce he was working on V11.3.
>>
>>36390436
https://desuarchive.org/mlp/thread/35100128/#35114433
>hey 15 i'm just a little curious, who is your favorite pony and why?
>Easily Twilight – I see a lot of myself in her character. Wasn’t a big fan of that last episode, though.
>>
>>36390783
https://u.smutty.horse/lyzvxwncssh.wav
>>
>>36391044
Then who's providing all the transcriptions? I don't think there are a lot of places like /mlp/ (that is, autistic enough to extract, clean up and transcribe hundreds of hours of voice lines).
>>
>>36390633
>cp77.txt
I do not have this file (or maybe my computer isn't able to search for it)
>cd Desktop
"The system cannot find the path specified"

The original game file is placed in:
G:\Program Files (x86)\GOG Galaxy\Games\Cyberpunk 2077\archive\pc\content
The extracted files are save in (raw):
R:\_CYPU77\TEXT\base\localization\en-us\subtitles\quest\panam
and extracted in:
R:\_CYPU77\TEXTout

So far I was desperate enough to get in contact with the cyberpunk77 modding community on discord, sadly they didn't had any answers either.

Well, worse case scenario I will need to wait a year for the mod tools to update to degree that will be actually useful in extracting the code from game.
>>
>>36391181
>I do not have this file (or maybe my computer isn't able to search for it)
I was giving you instructions to create it so I could see the file extensions in your directory. I don't know what files you have so I can't know where to look unless you use the tree command and share the results with me, or take a LOT of screenshots of every folder.
>G:
>R:
Wow you're using some unusual drive letters. So I'll try to explain this again now that I know.

Open command prompt and type "cd %USERPROFILE%\Desktop" and press enter.

Type (this time INCLUDING the quotes) tree /f "G:\Program Files (x86)\GOG Galaxy\Games\Cyberpunk 2077" > cp77.txt

Press enter and a file called cp77.txt should appear on your desktop. Upload that to smutty.horse.
>>
>>36391291
https://u.smutty.horse/lyzwtipbmfb.txt
Sorry, im not a clever guy, I just a simple man that wants to hear Panam and other CP77 character talk and stuff (and to share all the data here so others can create their won audio shitposts at their heart desires)
>>
>>36391442
Hm those /g/ threads were right, there are a lot of .archive files in there.
How much space does the entire Cyberpunk 2077 folder take up, and how much space does the archive folder take up (the archive folder inside the Cyberpunk folder)? Based on the filenames, I'm 99% certain the audio is either in archive\pc\content\audio_1_general.archive or audio_2_soundbanks.archive. I think these are likely encrypted and .dll files that run with the game are able to extract them.
>>
File: big boy.png (54 KB, 684x669)
54 KB
54 KB PNG
>>36391514
Cyberpunk 2077 folder: 59.8 GB (64,229,052,146 bytes)
archive folder: 58.9 GB (63,326,822,400 bytes)
lang_en_voice.archive file: 4.67 GB (5,016,571,904 bytes)
>>
>>36391555
cool trips. but yeah, audio_2_soundbanks is all the radio and music stuff, lang_en_voice contains voices and the file lang_en_text contains in game dialogue text that im trying to figure out right now.
>>
what the hell created this influx of new users?
it was at a cool 600 at most a few days ago
>>
>>36391672
Anon the entire TF2 community and /v/ have been all over it for the entire past week.
Just search "15.ai" on Twitter and you get hit with tens of thousands of results
>>
>>36391684
/v/ faded out of it like three days after it went up.
something else happened.
I'm suspecting something like with how Markiplier killed Artbreeder, and got them to finally patch that exploit that allowed infinite uploads.
>>
>>36391714
There are /v/ threads literally every day
>>
It's the return for error code 500: server errors.
>>
>>36391730
Working now
>>
>>36391726
they no longer last till bump limit
>>
File: file.png (1 KB, 265x52)
1 KB
1 KB PNG
lmao
>>
>>36389665 (Hacker Anon)
Thank you. The Step-Over part didn't work for me since it needed to step over several thousand instructions, and it still didn't get to the popup. I did this instead:
- Search for the string "is no longer supported in this version" using HxD. This gave me a hex string pattern I could search for in x64dbg. (For some reason, searching for the ascii string didn't work even though it's ascii in the binary.)
- Follow your steps to set up and run Animate.exe under x64dbg. In addition to unchecking "TLS Callbacks", I unchecked all the other boxes as well. Also whenever I ran into an exception, I went to the Exceptions tab in the same screen and clicked "Add Last" so x64dbg wouldn't break on the same exception again. I also deleted everything in the "Breakpoints" tab. I went through the process of loading and closing an FLA file while adding each exception to the Ignore list using "Add Last".
- Go to the Memory Maps tab. Right click anywhere and select Find Pattern. Paste the HxD pattern in. Double-click through each result until I find the one that matches the "ActionScript @0" string I see in HxD. (There was only one result when I copied the whole hex string.)
- Once I found the right memory location, right-click any line of the pattern in the bottom pane -> Breakpoint -> Hardware -> Access -> Word.
- Open any file the causes the popup. Now x64dbg will break on whatever is using the "ActionScript @0 is no longer supported" string.
- Open the Call Stack tab. Double-click through each of the lines that says "animate" in the Comment column. The first few are all doing some basic string operations, so ignore those. I ignored all of the ones that used "memcpy" or String-anything in the same function. The first one that's not doing a string operation is the one to delete. I followed your instructions to delete the call, test the change, and save the patched file.

Here's the patched Animate 2020 file:
- https://u.smutty.horse/lyzxuscmmxm.exe
>>
>>36391016
Based
Only a Twifag could be smart enough to pull this off
>>
>>36392075
>>
In case anyone else might need it: 19,261 lines of Geralt from The Witcher 3.

Transcript:
https://u.smutty.horse/lyzxurslast.txt

Data source:
https://www.nexusmods.com/witcher3/mods/3058/?tab=files

Text formatted according to 15.ai contribution guidelines.

Obviously I haven't manually validated the data given the scale of this dataset, but the transcript is curated to exclude instances of distorted audio mainly from The Last Wish (filtered out files with stereo tracks), cut out lines longer than 12 seconds (very few), and weed out some lines that didn't have an associated audio file or had muted audio.
>>
>>36392055
Yeah I knew that my way of doing it was inefficient, I just didn't know how to use any of the features to speed up the process. That was actually the first time I ever used a debugger, I tried editing the file with HxD at first but realized I'd only be modifying the ASCII messages instead of the popup code, so I downloaded x64dbg and tried to figure out how it worked.
>Open any file the causes the popup. Now x64dbg will break on whatever is using the "ActionScript @0 is no longer supported" string.
This is actually what I wanted the debugger to do from the start, I just didn't know how to implement it.
Thanks, I'll try keep all this in mind for the next time I attempt to use x64dbg. Did the "fl.showIdleMessage(false);" code do anything to stop the long running script message?
>>
>>36392243
Yup, that one's working perfectly too.Thanks a ton. I really should have googled that one before posting it.
As far as I know, there should be nothing else blocking us from processing the files in batch, other than my abysmal development speed.
>>
File: literal perfection.png (174 KB, 1280x1018)
174 KB
174 KB PNG
https://voca.ro/1hiA7sA46L44
>>
>>36392291
Sounds great. After a few attempts I was able to download the full 106GB of animation resources. I had to run unzip in the terminal manually for each archive because I didn't understand how to script it. I knew what needed to be done, I just didn't know how to write it. Things like that often slow me down.
I guess I'll try to watch some more Python tutorial videos on YouTube. I get frustrated if I have to watch a 6 hour video and end up learning nothing I can actually use. Starting out is the hardest part for me.
>>
>>36392332
If you already know your way around a command prompt, I'd recommend doing this to get started:
- Figure out how to run a basic Hello World program in Python. That's the fastest path to learning the bare minimum needed to run a Python script.
- Run this script: https://pastebin.com/QkGys3D7. You're probably going to use the "os" module a lot early on. That's an example of how to use it.

After that, run the python interpreter in a new command prompt window. Run:
>import os
Import the "os" module, which contains a lot of utility functions for doing things that you'd normally do in a command prompt.
>dir(os)
This will give you a list of all functions and variables in the os module. You can run dir() on any python object to get a sense of what you can do with the object. You can dump that information in a slightly nicer format with:
>print('\n'.join(dir(os)))
And finally, run:
>help(os.listdir)
You can run help() on any python object get read its documentation.
Eventually, you'll want to learn to use conditionals, loops, and functions. That should get you through simple scripts. If you don't like long tutorials, I recommend not watching them until you feel like you feel comfortable with the language. Tutorials, language references, API documentation, and style guides will fill in the gaps in your knowledge.
>>
https://u.smutty.horse/lyzzeltztsr.mp3
>>
>>36392617
Nice.
>>
had a double take when i saw this. a different kind of preservation, though
http://www.mlppreservationproject.com/
>>
>>36393272
this is boomer as fuck lol
>>
File: 1595623381779.jpg (174 KB, 1228x868)
174 KB
174 KB JPG
>>36393272
>>
>>36392617
I love this skit.
>>36393272
We should join forces. She could take lead on Pony Preservation Project: Toys (PPPT). Do you think she can learn to manufacture them?
>>
File: 4ever.png (150 KB, 474x250)
150 KB
150 KB PNG
>>36393272
god i'm old
>>
>>36393272
Man, I love internet rabbit holes like these
http://www.mlppreservationproject.com/links.html

I mean, look at this
https://the-fakie-hideout.weebly.com/
>"What the heck is a fakie?! "Fakie" is a term that MLP collectors use to refer to My Little Pony knock-offs."
>Last updated on October 8th, 2012.
>>
File: images (2) (1).jpg (21 KB, 739x415)
21 KB
21 KB JPG
https://voca.ro/1eOyUQcxCXhW

Wassup can a draconequus come up in your crib?
>>
base
>>
File: depfake.png (10 KB, 925x91)
10 KB
10 KB PNG
>>36393514
forgot image
>>
File: 614526.png (202 KB, 557x421)
202 KB
202 KB PNG
https://vocaroo.com/16UXY469o8t1
>>
>>36391672
Twitch mostly, a couple streams with 50k viewers have been flooding it when they stream.
>>
Two questions...

1) Why not pay to original artist or imitators to record samples of voices?

2) Did someone have a short tutorial to fake voices?
>>
>>36393881
>accepting payment to outmode yourself
i mean it's gonna happen anyways but the VAs probably don't want it to happen too quickly
>>
File: eqg.jpg (152 KB, 878x900)
152 KB
152 KB JPG
>>36393881
1) Even if the VA agreed to this, such samples would likely be almost useless. The amount of data is not nearly as important as the quality.

2) lurk moar, barbienigger
>>
>>36393881
barbie fuck kill yourself and lurk moar you fucking hog
>>
https://voca.ro/194pDz3KuuEI
>>
this was feelsy
>>36389621
>>
>>36393947
That's amazing, how did you do that?
>>
File: 1610039416438.jpg (117 KB, 739x739)
117 KB
117 KB JPG
>>36393935
>>36393937
>>
File: 2086962.jpg (49 KB, 564x563)
49 KB
49 KB JPG
https://u.smutty.horse/lzaffdkgoso.wav
>>
15 is fucking done
https://www.youtube.com/watch?v=xqJ-ujWvu7s
>duration and pitch of individual letters for extra control
>>
>>36394440
o shit
>>
>>36394440
>done
Looks more like somebody is pioneering the next step when it comes to being able to post process or manipulate the models to have more control on inflection, but the models themselves are extremely low quality.

Probably will be like the ngroks where we pioneered into making the voices have more emotion before 15 followed with much higher quality models and a bit more control.
>>
>>36394440
>duration and pitch of individual letters for extra control
We could've had that ages ago.I just have no idea how to write a GUI for it.
>>
>>36394440
can I train my own voice with this? no? than I dont really care.
>>
>>36394440
As is it's obviously worse than 15, but I can't wait to see what it'll look like in a few months of development.
>>
>>36394499
https://github.com/DanRuta/xVA-Synth/
>This is an Electron UI wrapped around inference of FastPitch models trained on voice data from video games.
You can just grab the Nvidia FastPitch and WaveGlow repos and make your own very easily. The creator of this tool only made the GUI.
>>
>>36394440
I'm seconding >>36394461, I'm guessing 15's gonna do something similar eventually but with his HQ models. He's already shown interest in similar things before: https://desuarchive.org/mlp/thread/36104106/#q36134679
>I do plan to make the pitch and rhythm contours controllable at some point.

>>36394490
>I just have no idea how to write a GUI for it.
I figured it would be the next logical step after getting the voice quality down. I think the best way to do it would be to have a grid where X axis would be time and Y axis would be pitch, and one could just drag syllables around however they need.
If there was a tool where doing that actually resynthesized the audio and didn't just time/pitch stretch it, that would be fucking awesome.
>>
>>36394201
i type the letters into 15 dot com and make flutteshy say naughty things yes
>>
>>36394440
I took a quick look at the GitHub repo and it looks like this is using FastPitch, which as far as I'm aware performs worse than default Tacotron2 in terms of naturalness, so this seems like old news to me.
Speech controllability is best done using an external audio file to serve as a pitch/rhythm contour anyway. It's on the agenda, and I'll probably tackle that in the coming months.
>>
File: file.png (2 KB, 257x47)
2 KB
2 KB PNG
holy shit these numbers can't be right?
>>
>>36394710
The queued counter should be correct. The servers went down for 20 minutes while I was getting coffee. Restarting now.
>>
File: file.png (1 KB, 220x37)
1 KB
1 KB PNG
>>36394710
yes
>>
>>36394490
If you create a model with the right APIs to support it, someone here will write the GUI. If no one else does, I can do it, or I can teach you how to do it.
>>
>>36394830
Teach me
>>
>>36394846
OK, but after work. Can you explain:
- What kinds of programming you've done in the past.
- What you know about how a browser works.
>>
>just been going through all the characters having them say "What a fucking waste of money"
>been laughing like a retard for the last 2 hours
Bless you 15
>>
File: all me.png (26 KB, 564x485)
26 KB
26 KB PNG
>>36394897
oh nvm then, the only experience I have is compiling stuff on github and trolling with inspect element.
come to think of it, i don't even know why i said anything.
>>
>>36393516
semantics. this is the term people know it by and it instantly explains what it is.
that said, I do hate normies, clickbait, and any modern terminology.
>>
>>36393520
Molestia you can't say that...
>>
I'm glad there are multiple projects, I just hope they share data to improve each other
>>
File: 2021.01.12_00.58.51.png (20 KB, 704x222)
20 KB
20 KB PNG
Does anyone know what's in these archives? I can't see the file list from just the first few pieces, and I don't feel like buying an actual fucking hard drive just to download them.
>>
>>36394763
I'm unable to put sentences into Text2. It doesn't seem to accept more than one word. Is this unintended?
>>
>>36395625
refresh your browser, it wasn't working in the morning but it's been fixed
>>
>>36390027
https://u.smutty.horse/lzahzawwsod.mp3
It wasnt worth i5
>>
>>36395772
>https://u.smutty.horse/lzahzawwsod.mp3
Background music is a bit too loud.
>>
File: Crusader Approved.gif (3.05 MB, 640x360)
3.05 MB
3.05 MB GIF
>>36395772
Absolutely and unequivocally based.

Kinda reminds me of SnoopyAnon's style
>>
File: 1609909382946.png (53 KB, 374x451)
53 KB
53 KB PNG
>>36395772
>applejack walking by rambling about sneed
>>
File: LyraFace.png (183 KB, 500x549)
183 KB
183 KB PNG
derpy.me/23EUs
Good Poni Content folder has just been updated with over 50 new pieces of content. It should now contain all significant content posted in the threads up to now.
>>
File: assets.png (297 KB, 1257x676)
297 KB
297 KB PNG
>>36395606
I just finished extracting that folder earlier today. It's all animation assets. There are a few episodes in there in .flv format. There are flash puppets for every pony so you can pose them. Here you can see me playing around with assets.
The background is from S9 E25 and there's Rarity in an unused bridal gown. I put Twilight in that chair and yes I deleted her wings. Now I just need to figure out how to use this stuff.

Pretty much anything you would need to make your own mlp episode is in there. The extracted folder is almost 200GB.
>>
Possibly played with the Scout Voice Actor on TF2 today (ChilofDestiny) he referenced his twitch in chat and started talking, I lost my mic audio but I was showing him 15.ai clips and he got a decent chuckle and later in the clip he uses 15.ai. Looked at his steam profile page and a lot of people said he was an impersonator but it was still a fun moment.

https://files.catbox.moe/4uo3b2.mp4
>>
>>36396062
So it's mostly .fla files? Can you post a screenshot of the folder? I was afraid of downloading it in case it's just retarded 30 gigabyte bluray rips of episodes or something. I'm curious to see what's been leaked/posted/made since those puppets back in 2012-2013.
>>
File: file.png (1.5 MB, 1463x827)
1.5 MB
1.5 MB PNG
>>36396092
>>
>>36396092
https://clips.twitch.tv/DistinctTenuousOkapiAMPEnergy
>>
File: files.png (134 KB, 1366x695)
134 KB
134 KB PNG
>>36396102
No it's only flash files. Backgrounds, characters, things like that.
>>
>>36396159
Huh. Alright, well I guess it's time to rip out a laptop drive and sacrifice my movie collection or something. I wonder why it's several hundred gigabytes, vector data is one of the easiest things to compress.
>>
SFM animators are already making hilarious skits using 15.ai
https://www.reddit.com/r/tf2/comments/kuty8w/busted/
>inb4 le reddit xD
>>
>>36396170
There's a 131GB scenes folder including 13 episodes. I checked the folder for The Last Roundup and every scene was there, so you could redo that episode entirely and swap in different characters if you chose to. The Ending of the End (1 and 2) take up 66GB and have multiple scene revisions, and every .flv file contains all the data needed for the scene, including audio so you can see how that would take up space fast. If we had access to editing every episode, all the data would add up to at least 2TB.
>>
>>36395772
https://u.smutty.horse/lzajbitivzz.wav
>>
File: png.png (629 B, 117x21)
629 B
629 B PNG
this can't possibly be accurate
>>
>>36396243
If it's raw audio then yeah, that alone would take up a significant amount of space. The actual files involved in production no doubt even more - concept art, After Effects projects, compositing, editing, etc.

>redo that episode entirely and swap in different characters if you chose to
That's sort of the idea, though I mostly just need them for reference and studying. The end goal is to have an algorithm that can piece together a scene by overlaying 2D puppets on top of 3D animation data, and make episodes from scratch.
>>
Hey anons, probably a dumb question but what software do you guys use to mess with audio stuff? I wanna make some stuff but I don't know where to begin with
>>
>>36396330
To my understanding most people start with Audacity as it's free, easy to use, and has just about enough tools and FX to spice up audios. Personally, I really like Ableton for audio work as its workflow is fast and its plugins are useful and versatile, but it's a little less free. Currently I run Logic Pro X for most of my stuff but it's restricted to MacOS and is mostly geared towards music production so not really worth the hassle IMO if you're just going for audio plays.
>probably a dumb question
Anon, you're not even in the top 100 for these threads.
>>
>>36396373
Thanks anon, will download Audacity and fuck with it to see if I make something
>>
Anyone else getting an endpoint not found on 15.ai?
>>
>>36396455
Yeah. Means either the servers went down from a new influx of traffic, or 15's uploading a new version, if not 11.3.
>>
https://u.smutty.horse/lzakggmlaaw.mp3
>>
File: finger ring friends.png (327 KB, 500x281)
327 KB
327 KB PNG
https://u.smutty.horse/lzakyvxrohu.mp3
>>
>>36397038
Nicely done.
>>
>>36395606
>Does anyone know what's in these archives?
I have some idea. It's all of the leaked FLA and SWF files we know of, plus some fan animation assets.

>>36389530
>>36381944
- [In progress] Run this on all files, all symbols. Adobe Animate shows a pop-up when loading old FLA files, and I'm not sure if I can automate that away with JSFL. I might need to set up an autoclicker to deal with that.
- ... [Done] Sample a bunch of files and make sure I can dump all of their symbols.
- ... [Done] Use the library name to identify & deduplicate symbols.
- ... [Done] Get rid of that popup dialog when loading old FLA files. Hacker Anon figure out how to do this. The patched version of Adobe Animate is in >>36392055.
- ... ... [In progress] Add Hacker Anon's steps to the Master Doc so we don't lose track of how to do this.
- ... [In progress] Update the scripts to open each file and dump its symbols one at a time.
- ... [ ] Run the script for however many days it takes.
>>
>>36396062
Shit, I still didn't bought the new disk and can't download it.
Nice work!
>>
>>36391555
So >>36391567 is right, I think the best luck you would have in extracting the .archive files is to run the game with a debugger and find out what .dll files extract the soundtrack, because they appear to be encrypted. Sorry I couldn't be of much help, I'd probably be able to put something together if I owned the game and could experiment with it myself.
>>
>>36397337
- [Done] Add Hacker Anon's steps to the Master Doc so we don't lose track of how to do this. https://docs.google.com/document/d/1xe1Clvdg6EFFDtIkkFwT-NPLRDPvkV4G675SUKjxVRU/edit#heading=h.8qg56l92l4ya
- [In progress] Update the scripts to open each file and dump its symbols one at a time.
- [ ] Run the script for however many days it takes.

>>36397405
You can start running the scripts without downloading everything. Here are some sample files:
https://u.smutty.horse/lyytnfcpiql.fla
https://u.smutty.horse/lyytnxavsma.fla
I'm almost done getting the scripts set up for a mass data dump, but it would be good to have a second person that knows how to do this stuff.
>>
>>36392243
For future reference: I don't know which debugger you used, but you can set conditional breakpoints for when EIP is equal to a certain value (e.g. address of the popup function, but this assumes that you know what the popup's function offset is) or if some string's address is being pushed onto the stack. The easiest way to root out these dialogs is get the strings output of the executable (like the trial message) and try to find what that thing's offset is in the rdata segment. That way you can set the breakpoint to trigger when something gets pushed with the value equal to that address.
>>
File: Do it for her.jpg (37 KB, 582x374)
37 KB
37 KB JPG
>>36389621
Just seeing this, feels crazy to listen to her voice and it being that convincing. feels like shes actually there. cant wait to hear more Luna stuff when its made.
>>
>>36397337
>the hacker collective known as /mlp/
>>
>>36396177
>no pony
>>
>>36397486
I forgot I said this earlier because you asked me for a detailed explanation on how I removed the ActionScript popup, but once you find out what you have to do in x64dbg, no one else needs to use the debugger in the future. Once you find out which bytes cause the ActionScript popup to appear, you only need HxD to patch the file.
I described how to do that here >>36389246
>After installing HxD, Run it as administrator if you're editing Animate directly from C:\Program Files\ as you need administrator rights to save there.
>Open Animate.exe with HxD
>Ctrl + F to open Find, set the tab to Hex-values. In the search field, put E8DAAC68FF. E8 should start at the second column at 00F9B8C0, preceded by a single D2. If that's correct, highlight E8 DA AC 68 FF, right click and select fill section. Fill all of that with 90. Save the file and run it.

I noticed that the instructions are on the Google doc. Even if using a debugger is the slower way, you might want to leave the steps there if they're ever needed, as editing hex with HxD is faster but is application specific and you don't want to forget how it was done.

Did you save the hex values you changed so you can create a guide to do this for Animate 2020?
>>
https://voca.ro/1jsJNKOicgfF
>>
>>36396724
Nice.
>>
https://u.smutty.horse/lyzhzdtnlvn.wav
>>
https://vocaroo.com/118k2BzykXv7
>>
>>36399011
https://voca.ro/1cDLIW59zdMC
>>
>>36399011
>>36399025
>>36399565
https://voca.ro/17fi3CrMeBna
>>
>>36399695
>>36399565
>>36399011
https://voca.ro/1bomWPB2mEvk
>>
https://voca.ro/1OikUqD0Krnt
>>
>>36399955
Holly molly, checked!

>Like they did
Anon, I... I did cam for feels, and the audio was sad enough, but this ending!
Well played I have the sad now.
Can I fix her? Please?
>>
>>36399885
tehehehe
>>36399695
this does kind of illustrate how the AI still can't replicate his highs and lows
>>
>>36399955
it's depressing sometimes knowing I'll never really be comfortable with technology as it changes and fucks over everything I ever cared about. I'll just have more and more trouble running old programs, then my old files wont be easily displayed anymore.. all the memories paved over.
>>
>>36400031
At some point it should be possible to just ask the AI to convert the data into a usable format.
>>
>>36400122
right? why is that shit lagging. where are the labor robots? it's been over a century since the term was invented, and its literal meaning is laborer..
>>
>>36400031
Can't you find / use open source program to open / convert them?
Open source program are most of the time not as good as paid one, but I always use them because even in ten years I will be able to open my files.
As for old programs, try to use a VM. The technology is almost mature to be usable now.
>>
>>36400225
it had better be
thankfully, even back in the day, I hated quicktime and shockwave stuff.
>>
File: car production.jpg (351 KB, 1280x720)
351 KB
351 KB JPG
>>36400215
How do you think they manage to pump car and stupid objects that cheaply?

>>36400241
After struggling a lot with some tricky old games, I will try and make a win98 / XP VM.
I mean, I could use some real hardware, but a VM have a lot of advantages, on being that I could theoretically take control of it from almost everywhere, even my phone, if I have enough internet.
>>
>>36400260
I mean yeah, okay, car assembly. but why ONLY car assembly?
hopefully if the industry realizes we don't need a new model every year and nobody wants a car, those factories get repurposed.
>>
>>36400031
I remember when in 2012 I had this awesome webcam app that allowed me do shit ton of cool stuff but after a forceful update some functions were removed, and with every update after that the app was able to do less and less until now it ever barely function at all.
Its like reverse Moore's law.
>>
>>36400274
skype, msn messenger, flash.. everything good gets sabotaged then destroyed. and crankpots keep passing new laws that make websites somehow culpable for everything we do, so they have to start tearing down pastebin, yahoo groups, smackjeeves..
>>
>>36400270
m8, you have no idea...
Everything is robotized in industry today!
I mean for medium to big company.
Food industry in particular is heavily robotized.
It's just not human shaped robots.
https://www.youtube.com/watch?v=uCgnWqoP4MM

>>36400274
>>36400296
Yeah, it's sad that old working software either didn't work any more or lost all it's functionality.
But computer usage changed too.
Back in the day, people used computer to create. Dam, even the windows default recorder have shitty effects like reverse or speed tweak. And now? It's a single button! Like, really, you tap record, the button change to stop, you tap stop and you choose where to save.
The default is not creativity any more, but normie ease of usage.
Simply because most of the remaining computer users don't know how the computer work.
And so company prefer to ship ironed out simple programs, and remove advanced functions because it may cause bug and are rarely used anyway.
But as a kid, it's because there were this weird "speed doubler" and "echo" on the micropone (pun unintended) windows program that I started creating audio clips.
Hell, you could only add soundfiles back to back , and to create radio effect, I used to save as heavily compressed µlaw wav.
I used everything that the program offers to me, and even more (and I still use pain from windows XP that I carefully copy from pc to pc).
Politicians are stupid and tech illiterate, and you can blame them for a lot of thing.
But the dumbness of today's software is mostly caused by companies and dumb users that only want to use google to watch netflix [sig].
>>
We were living in the wild west of the internet. Where you could go anywhere, post anything, DO anything, with little consequence. A few people obviously abused this, but for the most part, it really helped the internet grow into what it is. And the companies, small a big, supported this.

But now that not only does the government want more control via legislation, big tech wants more control because data is valuable and an open, untrackable ecosystem is antithetical to that.

So the internet has been, and now at a greater rate, will be locked down further and further.
>>
>>36400383
howcome I'm always seeing miserable people in hairnets doing shit like this by hand then
>the dumbness of today's software is mostly caused by companies and dumb users that only want to use google to watch netflix
tell me abou tit. my mom just got a new laptop, I'm like 'finally, a real computer. now I can teach her this and this and..' and it's a chromebook. she said she just wants to use it for netflix and 'apps'
>>
>>36400447
I dont understand it either, why is my data useful to anyone? I mean if you want to sell me products, you can HAVE records of my purchases. By all means, send me coupons and shit. That's not anything new. But what use is my browsing data except to fuck with me? How does it make them one red cent?
>>
File: PONY ZONE Thumb.png (3.52 MB, 1280x720)
3.52 MB
3.52 MB PNG
https://www.youtube.com/watch?v=BdkslodrGFc
Pony Zone, Lyra & Bon Bon Edition.
Tried some new things with this, not all of them worked. But, live and learn and all that.
>>
>>36400489
Data let's companies put up walls and barriers to control the experiences of it's target audience. The better data, the closer you can wall them in without your audience jumping ship, the more control you can have in what sort of environment your customers are in, and if you can completely control that environment, you can completely control what they see.

As an example, youtube ads. At first, on the side of the page. Then a box at the bottom. Then one, unskippable ad. Then multiple ads throughout a video. And now two, unskippable ads. Youtube has created a content platform, a box, restrictive enough to have it's advertising audience right where it wants them. It's collected enough data to say that a majority of users are tech illiterate enough to not only not block ads, but sit through multiple of them. They are finding ways to deplatform and remove content that doesn't bring in enough of a brain-dead audience for them to collect data and sell shit to.

This isn't some sort of orwellian sci-fi mind control thing. Companies want to sell you things, and the best way for them to do that efficiently is for them to have you right where they want you. More data lets them do that to more people with less effort.

Update your hosts list is a good START to stop data tracking:
https://someonewhocares.org/hosts/
Most ad-blockers subscribe to this list, but having it on your computer prevents ANY program on your computer making a connection to those sites, including viyda.
>>
>>36400564
But sales worked a lot better BEFORE all this nonsense. Nobody pays attention to ads or buys the shit in them. They're supposed to be figuring out what we want, and we can't want what we want if we aren't living how we want.
Flat fact is that business as a whole is doing terribly, and it does worse the more crap like this they pull. This is the antithesis of profit. This is "Dick Dastardly is winning the race but he stops to cheat anyway" shit.
>>
>>36400564
>Companies want to sell you things
I have literally never met anyone that has actually seen an ad and be like "yep gotta get that"
Either they already used the product, or they used a competing product already and are locked in.
The especially stupid ads are the food and drink ads.
Like pepsi and coke, its fucking retarded. Everyone who is going to drink them is already drinking them.
>>
>>36400509
Nice, I like how musical the dialogue is in this one.
>>
>>36400509
Would you look at that, it's nut o'clock.
>>
>>36400509
Amazing
We need more high-effort content like this in here
>>
>>36400509
Fantastic, though YouTube compression fucked the vocals, at least for me. It’s fine in the download but the video seems to make the instrumental overpower the pony voices.
>>
Server errors?
>>
File: 1605685398934.jpg (40 KB, 573x738)
40 KB
40 KB JPG
>>36400509
its time
>>
>>36401046
Yep. I've been experiencing them too. I think the people are stressing the servers
>>
>>36401059
Aw, that sucks. I was in the middle of an audio project but I guess I'll stop for now so nothing too bad happens to the servers.
>>
>>36401046
Doing some backend optimizations. Should be back in a minute.
>>
>>36401068
The timing on that is immaculate, I literally just got something back!
>>
blob:https://15.ai/0482d80b-3c48-4bb3-ab9e-d200d0bf7eab
>>
>>36401157
Anon, download the audio file and upload it to vocaroo, 15.ai doesn't host the blob file
>>
15.ai is kill?
>>
>>36401195
From 15's Twitter:

"The site will be down for ~1 hour while the model is updated from v11.2.5 to v11.3.1."
>>
>>36401197
Understood!
>>
>>36401193
thanks, but it seems God hates a retard like me
>>
Site's back up, updated to 11.3.1: https://15.ai/
>>
>>36401195
>>36401197
It's back now
https://voca.ro/1lnN8QjDSDzB
Seems like a good improvement! Pretty much zero noise in Sunset now on the first generation. It used to take several tries before I got a sentence with that little noise.
>>
>>36401374
Nice!
>>
>>36401157
https://voca.ro/1dAaEID12lW8
>>
>>36397899
Good idea. I didn't save the value, but I can get it again tomorrow and update the doc.
>>
>>36400600
It's not just about the amount of money they make. It's about the amount of money they make relative to costs. Collecting and storing an extra chunk of data on you costs them approximately nothing. If it makes them even a few pennies, their return on investment is through the roof.
Sales worked better before SEO was a thing. There's a back-and-forth between SEO and advertisers. Advertisers charge companies more if their ads seem less relevant to the users they want to display their ads to, so a market developed around basically hacking the advertisement algorithms to get cheaper ads. That also made ads less effective.
Advertisements are auctioned, so advertisers naturally adjust prices based on how many ads they need to show. If throwing money into advertisement was an easy way to get more profit, companies would fight each over the auction until it was barely profitable. That means they're willing to target people with worse matches for the ads until it's barely profitable to do so. For a lot of products, the cost of throwing up an ad is tiny compared to the value of getting or retaining a customer.
The way advertisers convince companies to use their platform is by saying they offer the best personalization so companies can spend less money targetting very specific niches. That means advertisers compete with each other by collecting more data and knowing the most about their users.
Companies like Coke and Pepsi put up ads to keep the product fresh on your mind. If you haven't had a coke in a few weeks, then the ad might get you to start buying again. Keeping the product fresh on your mind is also why companies like having famous people, references, or songs in their ads.
>>
>>36400509
Why do I hear Twilight sometimes?
>>
>>36401727
thankfully, recent trials done by big name companies where they cut back hundreds of millions in advertising resulted in no changes to their sales figures. they’re getting more success out of influencers peddling their wares. hopefully this leads to less data gathered overall by analytics companies.
of course, there will always be exceptions for companies that actually do use that data for other than sales, such as insurance and political campaigning. it could also be that those entities end up driving the demand for data enough that we see no changes in how data is collected... which would be awful.
>>
>>36401948
>>36401727
'ok'
>>
https://voca.ro/10o2h7yOl496
>>
File: 1610539346142.jpg (66 KB, 640x433)
66 KB
66 KB JPG
>>36401727
>>36401948
>>
>>36400509
https://u.smutty.horse/lzbabjbejdf.wav
>>
>>36401727
>>36401948
wtf
>>
https://u.smutty.horse/lzbatyxajip.mp3
>>
>>36402810
https://files.catbox.moe/4x4zv1.mp4
>>
>>36402810
kek
>>
File: roger p1.png (41 KB, 790x646)
41 KB
41 KB PNG
I've got some time off today so I felt like it was good day to start cleaning the transcripts of Mr Roger Interview, I have 25% of part 1 done, from the audio it's pretty clear that the interviewer was letting Roger monologue for most of the time so I feel like I can squeeze 18 minutes of audio out of each 30 minutes segments, so in turn it will hopefully be about two and half hour of pure clips.
>>
>>36403084
>Mr. Roger in Equestria
gonna be kino
>>
>>36403090
Yep, gonna look forward for whatever peoples creativity takes them.
My internet is acting bit funny so I will probably not able to train this model myself, so I will dump whatever the final number of clips, transcription text and raw audacity here.
>>
This whole thing is absolutely kino. Keep up the good work guys!
>>
https://www.reddit.com/r/mylittlepony/comments/kwbf5c/this_is_it_it_is_done_fully_animated_with/
>>
>>36403835
why would you...shill your work like this...reddit? oc?....please anon, you must try harder.....
>>
>>36403835
>the bot is more emotive than the human.
Kek
>>
>>36403853
this kek
>>
>>36400724
>>36400891
>>36400923
>>36400995
>>36401050
>>36402539
Thanks, glad you all enjoyed it!

>>36401912
She has one line at the very end, other than that I'm not sure. Is there any other spots you're hearing her?

>>36400995
Strange. I'll have to look into what Youtube compression actually does to audio so maybe I can prevent it from happening next time.
>>
>>36403843
It's not my work, I found it on le reddit and made me chuckle for >>36403853 this reason
>>
File: RAGE.jpg (46 KB, 750x728)
46 KB
46 KB JPG
>>36403835
>generic dark oc
>alicorn princess
>7th element
>janky linear pose to pose animation
>mcdonalds drive thru mic
>she's getting PRAISED for this
This is making me irrationally angry.
>>
>>36400509
Great work on all accounts, the music and dialogue compliment each other especially well. The Pony Zone is one of my favourite things to come out of this project and it's great to see you continuing with it.
One thing I will say to improve future works like this, try to focus on the characters themselves a bit more. I think most of the lines were rather generic and could have been said by any of the ponies. If you were to show me just the script with all the names redacted, I don’t think I’d have been able to guess which ponies were supposed to be starring. Admittedly not the easiest task with Lyra and Bon Bon, but there's still plenty of fan things you could reference to make the dialogue more interesting.
>>
>>36403853
lmao
>>
>>36403853

If this was a turing test, we would lose badly
>>
>>36403853
It's supposed to be a british accent!
>>
>>36400509
Oh god I need a Cadance/Chrysalis one.
>>
>>36404029
>Great work on all accounts, the music and dialogue compliment each other especially well.
Thanks mate! The instrument choices were almost entirely fueled by me just fucking around so I'm glad it still turned out okay.

>try to focus on the characters themselves a bit more.
This is definitely something I want to work on for future Pony Zones. The originals retain a lot of the characters personality which I feel make them a lot more entertaining. That will unfortunately probably get progressively harder as the characters get more obscure and made worse by me being shit at writing dialogue. Taking from fan stuff will probably be necessary for some characters.

>there's still plenty of fan things you could reference
I almost leaned more into the hands obsession/fetish that people sometimes give to Lyra, but decided against it, I think because at the time I was trying to keep my references restricted to canon. The "Feel me up, touch me all over" line was intended as a vague allusion to it, but might've been TOO vague in hindsight.
As for Bon Bon, she doesn't really seem to have much to take from, even in fanon/headcanon realm, at least from what I've seen.
Either way, more research into the characters personalities and quotes, both canon and otherwise, is definitely something I want to incorporate into my workflow going forward.

And on a COMPLETELY unrelated note, does anyone happen to have a list of all the episodes in which Celestia and/or Luna feature prominently? ^:)

>>36404358
Hmm, I was originally planning on just having bughorse paired with several changelings that would be constantly changing between characters, but actually Cadance/Chrysalis would be a really good pairing too, now that you mention it, and also more in line with the format.
>>
File: medium.jpg (77 KB, 484x600)
77 KB
77 KB JPG
>>36400509
Amazing

https://www.youtube.com/watch?v=7MOiUGMlgv4
>>
File: 1609867788135.png (274 KB, 500x500)
274 KB
274 KB PNG
>>36404439
>the fucking cartoon sound when Bruce Willis' head turns
>>
>>36403835
Lost it at "You are the princess of understanding"
Seriously tho, as bad as it is, the tool we created helped to spark creativity to the creator.
Was it a good idea, that's an other debate...
And that's all that matters.
>>
>>36404468
i agree, seeing people try and make their own content with ai is a good thing, but i'm on /mlp/ and therefore my post has to be nothing but negative
no anon, fuck off, it was pure reddit cringe and no one here would ever agree with you. you need to go back.
>>
File: celestia-eps.png (71 KB, 1078x757)
71 KB
71 KB PNG
>>36404384
>does anyone happen to have a list of all the episodes in which Celestia and/or Luna feature prominently
The wiki has tables like this for all the major characters
https://mlp.fandom.com/wiki/Princess_Celestia#Appearances
https://mlp.fandom.com/wiki/Princess_Luna#Appearances
>>
>>36404439
Haha, brilliant.

>>36404503
Perfect, thanks. This'll definitely come in handy for future stuff.
>>
>>36403861
I personally thought I heard Starlight in the Lyra parts sometimes.
People often upload audio in 1080p at least and tell people to switch to that to avoid audio issues, but I've never been able to find a difference. But I'm not an audiofag, so maybe there's something to it.
>>
Question, for a 44hz characters, what are correct options for exporting the wav files in audacity?
>>
>>36401948
Can you provide any references for the trials? It sounds like a very complicated thing to measure. I'd like to understand their methodology.
>>
>>36404909
From what I've heard, audio on Youtube reaches it's max quality at 720p. That said I'm going off of hearsay and rumor that I've never properly researched.
I also uploaded the video as an mp4 without really properly checking the codecs, so it's entirely possible that I accidentally uploaded the video with a subpar configuration that the uncompressed audio downloads don't have.
>>
>>36405482
>From what I've heard, audio on Youtube reaches it's max quality at 720p
damn, is this why any video i get from yt has audio quality stuck at 44100?
>>
>>36396159
>>36395606
Could someone upload just the flash puppets folder in this?
>>
>>36396724
https://u.smutty.horse/lzbglrqnzku.mp3
>>
>>36405629
Export them how? Does this work for you:
- https://u.smutty.horse/lzbgnphxmyg.ana

I can create those files with Adobe Animate, but I can't seem to open them again.
Once we get labeled data on which symbols correspond to characters, I should be able to dump all the puppets. Until then, I have no way to differentiate between puppets and other symbols.
>>
File: scootscoot.png (1.08 MB, 1500x2049)
1.08 MB
1.08 MB PNG
If You Wanted To... You Could... (You x Adult Scootaloo) -------> https://u.smutty.horse/lzbhhbixflf.mp3 (5:30)

Alternate title: Scoot Scoot

Started working on this a little under two weeks ago, hearing just how good Scootaloo sounded, and decided to do an audio with her.

Thumbnail image was made by this anon: >>36364844
>>
>>36406137
>>
>>36397899
>>36401612
I updated the doc with the hex editor option. That's much better than downloading binaries from random anons.

>>36405822
>>36405629
I realized that you meant the Puppets folder. I'm uploading it now. I'll post a link when it's up.

>>36397486
- [Done] Update the scripts to open each file and dump its symbols one at a time.
- [In progress] Run the script for however many days it takes. It's running now. No issues so far.
>>
>>36405629
1 - Stillz: https://drive.google.com/file/d/1qzy4j2Xj0qGL-5TQMjR7Nd3lp7gOalxT/view?usp=sharing
2 - Puppets: https://drive.google.com/file/d/1kEShwoqK4tXfIUooPwI8i418mJ4A_nHQ/view?usp=sharing
3 - Backgrounds: https://drive.google.com/file/d/1zft4-MxdShESWs-A_YZyAuCw6AY1EZ5K/view?usp=sharing
>>
>>36397899
Here's another one: Animate doesn't seem to give options in JSFL for exporting PNGs with a maximum size, and it pops up a window saying that it can't create a bitmap larger than 8192x8192 when trying to do so. I don't see an easy way to figure out what size it's going to calculate for a symbol.
This one is more difficult to reproduce since there doesn't seem to be a way to do it from the UI. The UI automatically limits the exported image size to 8192. I also don't see an option for disabling this one either in the UI or through a JSFL API.
I patched Animate 2020 to not throw the popup. I'm adding the instructions to the doc now. If you want to try reproducing this for Animate 2019, I can post the full instructions. It would be way easier if you can work with my github repo ( https://github.com/synthbot-anon/synthanim ). Here's the relevant FLA file that causes the popup:
- https://drive.google.com/file/d/1sJqHDYQfjVyblxZkTIDZ4HdN0iDZ9DBb/view?usp=sharing
>>
>>36406369
I'm not sure I understand. You want me to get rid of an 8192x8192 popup in Animate 2019? Is it just a warning like the ActionScript popup or is blocking this from appearing going to cause render issues? Do you want me to get rid of the box and force Animate to scale all larger bitmaps to 8192x8192 because that's a little more difficult than blanking out a few instructions.
>>
>>36406137
This is really good, well done!
>>
File: Spoiler Image (228 KB, 600x600)
228 KB
228 KB PNG
>>36404490
You don't understand Anon.
/mlp is one of the last bastion of 4chan spirit.
The hate thing is a cancer that destroys us from the inside.
Look at the early days of 4chan, joyful meme, creation for the lol and nothing more, and calling others faggot just cause it's funny.
Other boards, like /pol, are spammed by degenerates whom think being edgy cringe "kys nigger" is both the way to act on 4chan and to live irl.
The reddit video is cringe, but just look at ponyzone, even if the realisation is way better, don't tell me that the subject is not cringe (just closer to our taste).
Going back? Kek, where? I'm here from so long that I'm practically part of the furnitures at this point. So kek[/spoiler] yourself flaming faggot, and dilate your pupils of joy by seeing colourful equines doing cute things. I hope you will have a nice day, nigger.
>>
>>36394440
>that Joshua Graham
Disgusting!
>>
>>36404029
Here's a partial dump of the animation symbols that we'll need labeled:
- https://u.smutty.horse/lzbigbtrrqg.7z
The way I'm dumping them, each FLA file gets its own folder. Within that folder is 1 image for each symbol that's actually used for the animation. I can't yet detect if a symbol actually needs to be rendered for the shot. Some of that might require manual work since there might not be an easier way to tell if, e.g., one symbol is obscuring another. The consequence right now is that some characters will look weird when, e.g., only the top half is in the shot, and the animators didn't bother animating the bottom half.

This is just so you can get a sense of the task and so we can start figuring out how to label this.
- Ultimately, what we need is to label each symbol with the AI model it should be fed to. So if we see Derpy's head, it should be labeled as "Derpy" since that data should be fed to the Derpy Animation AI. It might be useful to have additional labels, but this is the bare minimum that we need.
- We only need to label the top-level symbol. So if Derpy's full body and Derpy's left eye both show up in an FLA file, we don't need to bother labeling Derpy's left eye. I can automatically label that from knowing that the symbol corresponding to Derpy's full body is marked "Derpy". The symbols are hierarchical, so my scripts can see that Derpy's left eye is part of Derpy's full body.
- You'll need to keep the filename in-tact while labeling. That's my only way to link your labels back to the original symbol. It's probably fine to move files around to different folders. It'll cause a bit of extra work, but I don't expect it to be much.

If needed, I can create tools to help you label the files. Let me know if you have any thoughts on how to go about this.
>>
>>36406393
It won't cause rendering issues. It's just a popup like the ActionScript popup. Animate ends up not rendering the offending image, but my script can just skip it in that case. I've already done it for Animate 2020, so I'm good to go. This is just if you wanted to add the Animate 2019 instructions to the doc.
>>
File: 1582486902602.png (14 KB, 450x450)
14 KB
14 KB PNG
>>36406137
>Every time I tried to tell you how I felt, I chickened out
heh
>>
File: PewDiePieThumbsUp.gif (1.38 MB, 320x180)
1.38 MB
1.38 MB GIF
>>36406137
Very impressive stuff, as always.
>>
>>36406137
Your stuff is always on a whole other level
>>
>>36406416
not sure how to feel about this, on one hand anon, there will be ton of cringe shit made by edgy kids and autists, on the other hand those cringy animation are unintentionally hilarious so there is a good chance someone will make a pony version of the 'Food fight'.
>>
File: 1604181477079.png (640 KB, 600x528)
640 KB
640 KB PNG
>>36406137
>chickened out
max kek
>>
File: 1323018892805.png (80 KB, 603x424)
80 KB
80 KB PNG
>>36406137
this is too good
>>
>>36406137
Genuinely great work anon
>>
File: 1605168855295.gif (806 KB, 612x630)
806 KB
806 KB GIF
>>36406137
Amazing work, my dude. The future is now.
>>
>>36406137
This, but kid scootaloo
>>
https://voca.ro/1dbrl1CfYp19
Can't stop making trixie say movie references
>>
https://vocaroo.com/1d46dMuF5uwi
Derpy has an important announcement.
>>
File: fluttershy book.jpg (10 KB, 223x226)
10 KB
10 KB JPG
https://vocaroo.com/12qaw5RhMQ0w

The breezies have come
>they have come to make peace with anonymous
>Equestrias resident and only human
>In a long parade they march into Ponyville, looking happy and festive as they sing their breezy songs and wave their little white flags
>Finally, they line up before Anonymous
>Offering their tiny colorful gifts as they kneel in surrender
>Unfortunately for the breezies, peace was never an option for Anonymous
>He starts to flail his arms around, violently swatting breezies out of the air
>The breezies attempt to flee, but like ripe cherries their soft and fragile little bodies burst beneath the soles of anonymous big leather boots
>Several almost make it outside, but slamming shut the heavy oakenwood door makes a swift end to this final delusion
>And so, peace and quiet returns to Anonymous humble little horseland abode
>The end
>>
File: 7848.png (1.51 MB, 6346x4346)
1.51 MB
1.51 MB PNG
>>36407855
https://vocaroo.com/13N2wnwyRd75
>>
>>36407855
>>36408016
kek
I still can't believe how far we've come, it's like a fever dream
>>
>>36390030
>>36406464
>>
>>36408016
https://u.smutty.horse/lzboloycgfe.wav
>>
>>36400509
Can you upload the lyrics / script somewhere?
>>
>>36406822
>>36406849
>>36407141
>>36407494
>>36407510
Thanks, lads. These audios are always tedious and frustrating, yet fun to put together!

>>36407137
>>36406817
Funny enough, I had the same thought. I wanted her to pause mid-word and say "wussed out", like the original phrase bothered her even at this point, but it was difficult to get her to say it naturally even with splices. But I feel like she probably would've gotten used to the phrase by the time she was an adult, so I left it as is.
>>
>>36408150
I don't actually have the full lyrics written out anywhere. I can do that real quick if you need them for something.
>>
>>36408180
Yes please, I would appreciate it immensely.
>>
>>36408185
https://u.smutty.horse/lzbotixsmto.txt
There we are. It's also now in the video description.
>>
>>36408137
https://u.smutty.horse/lzbouheapcd.wav
>>
File: Screenshot.png (91 KB, 980x501)
91 KB
91 KB PNG
>>36406464
Not sure I fully understand what needs to be done, I think I may just be misinterpreting some part of your description. You say each symbol needs to be labelled with what it is for the AI, but from what I can tell every one of these pictures already has labels in their filenames similar to what you describe, like "AJs Head". Or is this just meant to be an example of what the labelled symbols should look like?

>>36406137
Great work as always, I don't think I can really find much to criticise so I'll just say thanks for continuing to show us how to get the best out of the AI.
>>
https://yuki.la/mlp/33700529
Looking back at the first PPP-related thread ever is pretty fascinating.
>>
>>36408402
There are consistency problems with the filenames, so we'll need to attach more standard labels to them. Also, some of the files labeled with a character's name contain more than just that character. For example, there are some images there marked DERP that contain both Derpy and a broken piece of wood.
If we wanted to start by training a Twilight Animation AI, at the least, we would need to figure out which symbols correspond to Twilight and just Twilight. One way to do that would be to find all of the symbols whose name contains TS, Twi, TwilightSparkle, etc., then remove all of the symbols that contain non-Twilight props/characters. Then repeat the process for each character/prop that we'll want to render & animate.
I can post examples later tonight if it's still not clear.
>>
>>36408503
Okay, that makes more sense. Presumably then we'd only want the symbols that contain exactly one particular thing, so maybe a good first step would be to go through it all and dump anything that breaks that criteria.

>I can post examples later tonight if it's still not clear.
More examples certainly won't hurt.
>>
>>36406240
Thank yo so much. Could you also upload the effects folder? and whichever folder has the fashion stuff shown here >>36396159 sorry if I'm asking for a lot, thanks again
>>
>>36409004
You're not asking a lot, and no problem. Clipper did the hard work of organizing these files.
The fashion stuff is in the Stillz folder here >>36406240.
4- Effects: https://drive.google.com/file/d/1_NkVB8q2lE9YXFi5EPuyuAVYLP8YCtb7/view?usp=sharing
>>
File: Dark timeline.png (124 KB, 1002x332)
124 KB
124 KB PNG
>>36408445
Meanwhile, in the dark timeline...
>>
>>36409222
if only...
>>
>>36409222
Could you imagine if that actually happened? If the thread died right then and there and no one decided to take the helm at collecting all the data. We wouldn't even be looking back at this thread right now.
If you think about it, it's actually a miracle that any of this is even happening right now. We could have easily fallen into this dark timeline if some random troll decided to reply just that. Just thinking about it makes me shudder.
>>
>>36409241
Could you imagine what we'd have if the MLP Animation threads didn't die? Why do some projects live while others don't?
>>
>>36409222
>2019
It feels like an eternity
>>
>>36409245
One of the things I've noticed after many projects is this:
Projects that start organically on /mlp/ generally succeed while ones that start offboard are generally doomed to failure. Doesn't matter if people want to take it here and make it happen, the momentum just doesn't come. An observation anyway.
>>
>>36409241
Anon stop, you’re gonna give me nightmares
>wakes up only to realize that it’s still 2019 and none of this AI stuff happened
>>
>>36409241
Yeah, I made that image as a joke but the more I think on it, the more freaky it is to me that the fate of the project very well could've rested on someone brushing it aside with a troll post. The fact that the project not only lives, but I would argue is thriving, and contributing to the progress of cutting edge voice synthesis tech is really amazing.
>>
>>36409290
It is fascinating how many things we take for granted are the result of various factors coming together at the right time in the right places.
>>
>>36409308
Imagine being able to see that all happen from a distance.
>>
Rome, I know you lurk here and I want to tell you that you are a fucking madlad for saving this trixie VA audio from years ago:
https://iwiftp.romesilvanus.io/Pony/Audio/Vocaroo/Trixie%20Lulamoon/Trixie%20Lulamoon.mp3
Really, thanks a lot man. I kind of waifu her now and this is great.

Back when I was in college 5 years ago, you know amateur VA's were on here recording Anon's posts right? I saw one in a thread.
When I heard this VA voice Trixie, uh, how do I say this? This is a bit crude but I... I had to try my hardest to not cum in my pants in a room full of people.
>>
>>36409612
that's not ai voice anon, she barley sounds like Trixie, you should use the ngrok and 15 to recreate that with proper voice.
>>
File: 1593721268676.jpg (209 KB, 580x594)
209 KB
209 KB JPG
>>36409629
its rome he archives everything anon
>>
>>36406137
What software do you use to edit audio?
>>
Is anyone else having a rough time generating audio on 15.ai? In the past 24 hours or so I've mostly received various server errors and the few occasions where it works it's been painfully slow
>>
>>36409629
>5/10 reading comprehension
I admit it's a tad unrelated to the thread, but I wanted to thank Rome for his service because he lurks here. I had the same thought of recreating her lines to see if it gives me the same rock hard poner.
>>
>>36410248
I've been on the phone with AWS support for a few hours now. It's a weird issue that we're still trying to fix.
>>
>>36410159
Fl Studio.
>>
>>36409612
Whose voice is that?
>>
>>36410357
Thanks!
>>
>>36410357
now i just need to find my eyepatch
>>
>>36408559
For episode 506: https://drive.google.com/file/d/1VNeJDaOsAjmUPzkWf9geuoQbRJd_NfMv/view?usp=sharing
- I first split the files into "unsorted" and "rejected". Every file that contained "char" in the filename was moved to unsorted, and everything else was moved to rejected.
- I then created a "sorted" folder. Inside the sorted folder, I created a sub-folder for individual characters. I went through each character in "unsorted" and moved ones I recognized to "sorted".

I see a problem here. A lot of the files have duplicate names. I'll need to batch rename them to include the clip they came from, that way there won't be ambiguity when we move files around from rejected to unsorted to sorted. I'll do that before uploading.
I also see that a lot of these symbols have character + props rather than just characters. It might be worthwhile to split the character folders into, e.g., "apple bloom_clean", "apple bloom_noisy", and so on.
I'm not certain of this, but I think we're also going to end up with a lot of unnamed characters. Maybe we can sort those into folders like "generic earth pony".
>>
File: 954800.png (1.78 MB, 9000x10000)
1.78 MB
1.78 MB PNG
>>36408445
>mfw third reply is mine
>>
>>36410561
Are you going to do this by stitching together pngs and training the ai on them?
>>
>>36410620
Yes, basically. I expect it to be done in three parts:
- Generate the pngs for the character parts, or figure out how to reuse existing ones if possible.
- Stitch together character part pngs into a character pose png.
- Stitch together character pose pngs into an animation.
>>
File: 7a6.jpg (35 KB, 600x754)
35 KB
35 KB JPG
>>36410271
Maybe it's just me but it seems to be working much better now... Thank you again for your hard work and dedication, you're the fuckin man
>>
File: 1596605219013.jpg (55 KB, 600x401)
55 KB
55 KB JPG
>glados: im just saying, arent you getting tired of saying the n-word
>demo: why is everyone looking at me?
>>
>>36410271
>AWS cancels 15 for not regulating calls to insurrection and violence on his platform
>>
File: Spoiler Image (226 KB, 1000x1000)
226 KB
226 KB JPG
>>36410970
Well, if he ever decides to build his own AWS Im sure he will get a lot of clients.
>>
File: 1599039443828.jpg (116 KB, 720x1280)
116 KB
116 KB JPG
>>36409612
>>
File: 6546.jpg (69 KB, 612x612)
69 KB
69 KB JPG
>>36411218
>>
>>36411218
I showed you my >hooves. Answer me.
>>
Found on Twitter

https://u.smutty.horse/lzbxehbhugn.mp4
>>
>>36411446
kek, wtf?
>>
bump, also im almost 2/9 done transcribing.
https://vocaroo.com/14IgMzNEGBZB
>>
https://twitter.com/EquestriaDaily/status/1350216292503080961?s=19
I'm honestly surprised it took Seth this long.
>>
Wish I had a time machine. I'd take all these voices back to 2012.
>>
>>36413070
I remember there was an AI complilation on EQD around the time LittleShy posted this video https://www.youtube.com/watch?v=XPMCR4uyaYU&ab_channel=LittleshyFiM
>>
>>36409241
It's depressing enough thinking of all the great things that could be happening but aren't because literally only so many things can happen
>>
>>36413859
That's why we need CelestAI.
>>
>>36414147
Don't worry, the USA is getting the same thing but instead of a large white alicorn voiced by Nicole Oliver it's a corrupt, racist, hateful vaguely brown woman named Camel Toe.
>>
>>36414152
based
>>
File: 1590302158498.png (315 KB, 1100x1400)
315 KB
315 KB PNG
https://www.vocaroo.com/1g9G4HG9Mavv
>>
>>36414152
oh so gildafags will be right at home then
>>
Been having fun with accents

Aussy Jack - https://vocaroo.com/1nEVXpmEVRQc
>>
File: 1594685103447.jpg (40 KB, 532x590)
40 KB
40 KB JPG
>>36414407
You gave me an idea and I can't believe I got it to work.

I have finally made Octavia speak Scottish.

https://www.vocaroo.com/18UoHiGHtuug
>>
how are you changing the accents
>>
>>36414407
In response to you're triumph,
Irish Pinkie - https://vocaroo.com/1esS14LjcE1J

And for everyone else

German Rarity - https://vocaroo.com/1jYhbwAwRG4T

Tried to do German Applejack and this happened haha - https://vocaroo.com/1lqwDYlNeixs

>>36414425
Figure it out my dude!
>>
>>36414440
Whoops, I replied to myself like an idiot! >>36414421
>>
>>36414440
based
>>
>>36414421
>>36414443
Nice. I still can't get over the fact though that "Fookin" sounds exactly how it's meant to sound. Almost like 15 added british slang and memes to the dictionary too. And that my transcription is working on mostly every other character.
https://www.vocaroo.com/1iXvKEJs6A8w
das somebahdy na know how ta flusha toyleh' when they've hahd a SHET?! ,, Well at was fookin wanav yaz! ... DISGUSTAAN!!!|Fuck you!
>>
>>36414468
fucking lol'd
>>
>>36414468
Have you also noticed that these days, accents all seem to be more muddled and soft? as if the fact that everyone is watching videos from everyone everywhere, but mostly america, is causing their phonemes to change a bit. Like I swear irishmen today sound like they're americans just saying irish accented shit written out.
>>
>>36414606
On the topic of Irish voices, what's your opinion of Genevieve O'Reilly in Overwatch (as Moira)? https://www.behindthevoiceactors.com/Genevieve-OReilly/
>>
Hey Cookie, this mighty interest you, mozilla has a website called Common Voice were some people read short transcripts and others validate it, from what they are saying they are making all the valid audio clips available for people to download.

https://commonvoice.mozilla.org/en/datasets
>What’s inside the Common Voice dataset?

>Each entry in the dataset consists of a unique MP3 and corresponding text file. Many of the 9,283 recorded hours in the dataset also include demographic metadata like age, sex, and accent that can help train the accuracy of speech recognition engines.

>The dataset currently consists of 7,335 validated hours in 60 languages, but we’re always adding more voices and languages. Take a look at our Languages page to request a language or start contributing.
>>
Where do I go to get sound effects and such? It's in the leaks, right? I don't have a link to it tho
>>
>>36415022
https://mega.nz/folder/0UhSmYAB#WBrB-qCprQTofkAhwMp5CQ/folder/MVggyATb
here, the link its somewhere in the OP google doc.
>>
>>36415032
Thanks. Now I just gotta learn audacity.
>>
File: pngegg.png (83 KB, 900x861)
83 KB
83 KB PNG
https://u.smutty.horse/lzcivaaunvq.wav
>>
File: 1347358.gif (787 KB, 507x508)
787 KB
787 KB GIF
haha
https://vocaroo.com/1R9KlTIPwZgG
>>
File: L-lewd.gif (725 KB, 245x172)
725 KB
725 KB GIF
>>36415791
Fuckin' nice
>>
For the record, this is /co/'s 15.ai thread:
>>>/co/120315296
>>
>>36415791
https://www.youtube.com/watch?v=YBfEDFOMiZU
>>
Text-to-image notebook: https://colab.research.google.com/drive/1FoHdqoqKntliaQKnMoNs3yn5EALqWtvP?usp=sharing#scrollTo=-irecVJr_Zc9
Maybe someone here can train it on pony images.
>>
>>36415902
>crosslinking to /co/ is still disabled
How long will Mods be butthurt?
>>
>>36416450
Crosslinking from /mlp/ to /co/ is specifically disabled, newfag.
>>
>>36416450
i can't believe anyone's actually this retarded
>>
>>36415902
It's pretty fun
https://vocaroo.com/1EWZjjzRJxZk
>>
>>36416500
This is just a parody of the TF2 one retweeted by 15, isn't it?
>>
https://vocaroo.com/1ja7jPHZWpmJ

Did I do it right?
>>
say i wanted to change my google maps voice to my waifu's, how would i do that?
>>
>>36416589
Waze has the option to use custom voice recordings for its directions. Generate suitable lines for your wAIfu of choice and feed them into that.
>>
>>36414652
seems like it varies tremendously, some words sound downright american, others are irish as fuck (like the way she says 'habit')
>>
>>36416500
this is fucking fantastic. Imagine once we can get the few little kinks ironed out
>>
File: SadLyra.png (95 KB, 600x670)
95 KB
95 KB PNG
https://u.smutty.horse/lzcmelkidyk.wav
Poor mare
Original post: https://desuarchive.org/mlp/thread/15644209/#15650690
>>
File: 532454.png (4 KB, 359x89)
4 KB
4 KB PNG
>>36417111
cursed comment
>>
File: angy.png (150 KB, 920x984)
150 KB
150 KB PNG
Accidentally put this in its own thread without realizing this project existed, really glad i'm caught up now. Here's some first attempts done using the 15.ai.
>tfw it's a crossover episode

https://vocaroo.com/17Iqtlt18rJO - part 1
https://vocaroo.com/162BFrAh2E7X - part 2
https://vocaroo.com/130u5HDi2iZL - part 3
https://vocaroo.com/17JHk6SJmKyO - part 4
https://vocaroo.com/13C2lKWh6gsV - part 5 final
>>
File: 1603770151977.jpg (27 KB, 207x199)
27 KB
27 KB JPG
>>36417135
>comment
Can you just fucking go back already.
>>
>>36417138
this is a wild fucking ride of attempts anon. not saying it's bad, just didn't know what I was getting into. pretty neat though, and you should keep up the good work.

the dash, scout, and demoman part was probably your weakest link here.
>>
>>36417184
if only the cadence and intonation were easier to edit. the parts that don't work are just a letdown after all the parts that do work
>>
>>36394830
>If you create a model with the right APIs to support it, someone here will write the GUI.
Alright, I'm close to the point where this might come in handy.
Can you describe;
- what the inputs for the model should look like to be convenient for you?
(e.g: should each letter be a dict with pitch and energy, or should it be like 3 lists, list 1 = each char, list 2 = each pitch, list 3 = each volume, and so on)
- Which inputs do you think you can make intuitive? I have pitch, volume, voiced flag, spectral tilt, pitch consistency and a couple other attributes that can be assigned on a per letter basis. I'm guessing for a first prototype we go with simple just pitch or just volume?
>>
>>36417111
>trips
Even in her saddest moments this mare can't stop stealing all the gets.
>>
>>36417486
Just like she can't stop stealing Bon Bon's fancy oats.
>>
>>36416398
I'm not at all versed in this stuff. So, could you explain real quick how to train a model with that? From those cells I can't spot any sort of input to feed it pony images with, so I'm not sure how it works.
>>
File: 1608015520909.png (1 KB, 230x37)
1 KB
1 KB PNG
wtf is going on? Somebody trying to script spam the site again?
>>
>>36417725
could be some big twitch streamer playing with voices again or amazon servers going full derp.
>>
So I've been able to get Spike to do some really great and deep moans on 15.ai. However, it takes patience. I use "Mmmhhh" and "nnnhhh" and try to contextualize it with single words such as "tired, calm, sexy". It takes quite a few generations to do it. You could also use "ohhh". The biggest note is not putting too many m's n's or h's. It may become slurred. Also, separating them with commas is a must.
>>
>>36417755
I think it was the latter, looks like it's fixed now.
>>
>>36416409
get 4chanx
>>
File: 1496633027258.jpg (12 KB, 261x193)
12 KB
12 KB JPG
>>36417757
...not even a spikefag but post examples
>>
File: 1595234885106.gif (949 KB, 563x450)
949 KB
949 KB GIF
https://www.vocaroo.com/17YdDui1IUHj
Just since every 15.ai thread has at least some version of this I might as well oblige
>>
File: 1574150324073.gif (1.04 MB, 720x720)
1.04 MB
1.04 MB GIF
>>36418035
Thx Chryssi
>>
File: 1610577808742.jpg (472 KB, 828x912)
472 KB
472 KB JPG
>>36409222
>>36409241
>One grumpy anon has the power to change the course of history for an entire board
*PFFFFFF*
>>
>>36418035
Fuck, that's a rollercoaster for sure.
>>
>>36418035
>no mane
That's where you're wrong, cheeselegs.
>>
>>36408173
>explanation for keeping the phrase
I love how you thought all the way through with that and took into account how she would have reacted and eventually settled with it over time.
>>
>>36417111
A lot of extra music at the end. What's happening during that time?
>>
File: AnonShrug.gif (3.03 MB, 800x450)
3.03 MB
3.03 MB GIF
>>36418401
i kept playing piano past the end and didn't cut it out
>>
>>36418032

https://voca.ro/1ogaa88heRkb

Here you go
>>
File: chrysalis.gif (2.28 MB, 503x360)
2.28 MB
2.28 MB GIF
>mfw fags keep making one sided romance clips

https://vocaroo.com/1i1ITYQzIMvx
>>
https://vocaroo.com/1hDLGC4vKrW5
>>
>>36418532
nice job on the effect. if only she'd said CONtent instead of conTENT
>>
>>36417408
Both inputs formats would be easy to handle. It's a difference of a couple lines of code. I think the per-character attributes would make the most sense. That way we can eventually support streaming TTS, if we ever get that far. So I would send a single list, and each item in the list would be a dict containing a character, pitch, volume, voiced flag, and so on.
Note that I can't work on this until I get the animation stuff in a state where it can run without me putting time into it.

Current plan:
- I'm writing a script to rename all of the animation files. I'll give the same script to Clipper so he can move files around by pattern-matching on the filenames. I'll run the script on what I currently have so he can move around files freely without killing my ability to link them back to their original symbols. I expect it to be 1-2 days before he can start organizing/labeling the symbols.
- While Clipper is organizing the symbols, I'll write a second JSFL script to dump all the data we'll need for creating animation AI. I expect that to take 1-2 weeks. Given that it's taking days to dump just a single frame from each symbol, I expect I'll need to run the script for somewhere between 4-12 weeks to dump all of the relevant data for each character. If any other anon sets up the dev environment for running JSFL scripts, we can split up the task by episode to speed things up.
- Maybe you can try to get an initial TTS model + API up with Colab instructions before I finish writing the second JSFL script (about 1-2 weeks from now). The model doesn't have to be good. It just has to be something I can develop against.
- While my animation-dump script is running, I can work on creating an interface for your API.
- I can upload animation data samples ASAP and intermittently while working on the TTS interface. While I'm doing that, maybe you can start looking into creating the animation AI models that would work with the animation data I dump.
>>
>>36418574
I only understood like half of that.
>>
File: 1607780539631.jpg (432 KB, 727x992)
432 KB
432 KB JPG
>>36406137
Wow... just wow, I'm not even into Scootaloo but just... wow...
>>
>>36418594
is there a reliable way to force the AI to put emphasis on different syllables?
>>
>>36418645
Yes. Read this: http://www.speech.cs.cmu.edu/cgi-bin/cmudict Read it again if you've already read it.
>>
>>36418653
where do you put the 0, 1, or 2?
>>
File: chrysalis.png (40 KB, 231x199)
40 KB
40 KB PNG
>>36418653
thanks bro, I must've overlooked that

here's the fixed version
>https://vocaroo.com/13PoHCXL7qx6
>>
>>36418604
Meanwhile, if other dev & AI anons want to help:

If you want to get in the weeds and start working on the second JSFL script directly:
- Follow the "Extracting FLA animation data" section here: https://docs.google.com/document/d/1xe1Clvdg6EFFDtIkkFwT-NPLRDPvkV4G675SUKjxVRU/edit#heading=h.j03ylpv5bp0r. The "Patching Adobe Animate" section is optional.
- You don't need to download all of the data. It's find to download just a couple FLA files and work with those.
- Ping me. I'll write out instructions on how Adobe Animate's data structures are laid out, how to access them, and what we'll need to dump.

If you want to help run the second JSFL script when the time comes:
- Follow the "Extracting FLA animation data" section here: https://docs.google.com/document/d/1xe1Clvdg6EFFDtIkkFwT-NPLRDPvkV4G675SUKjxVRU/edit#heading=h.j03ylpv5bp0r.
- The "Patching Adobe Animate" section is NOT optional. If it's too difficult, I can package my patched versions of Adobe Animate so you can run that.
- If you don't want to use Animate 2020, I (or maybe Hacker Anon) can patch the 2019 version here: https://mega.nz/folder/GRVFHKyL#_65VIygOZ0mnWr9-V_8zOA.

If you're interested in looking into creating AI models for the animation AI, post here. Our Animation AI plan in the Master Doc is currently outdated. I can update the section the describe what data we'll have and what AI we'll need. At the least, we'll need image generation AI for creating character parts, (probably) generative graph neural networks for piecing together character parts into poses, and sequence generation AI for animating the poses. If anyone sees papers doing something similar, please post them.
>>
>>36418675
0 is for no emphasis, 1 is for primary emphasis, and 2 is for secondary emphasis in a word with more than two syllables.
>>
>>36418708
So where do you put them?
Like if I'm writing out "I contest the outcome of that contest"
>>
>>36418766
I {K AH0 N T EH1 S T} the outcome of that {K AA1 N T EH0 S T}
>>
>>36390030
>>36408559
The first 10 episodes are here:
- https://drive.google.com/drive/folders/1qc36X85MfoRoAfFfnzbMIhCIeD_RpitD?usp=sharing
The remaining 3 will take a while. 922, 924, and 925 are about 3x as large as the other 10 combined.

I'm not sure if you've had more thoughts on sorting these over the last few days. It should work to throw them into these folders:
- rejected/ for symbols that aren't of characters
- unsorted/ for symbols that we might want to use but don't know how to label it
- sorted/clean for symbols with a full character and no props
- sorted/noisy for symbols we would can clean with little effort

I wrote a script to help move around files:
- https://drive.google.com/file/d/1-nbUydtioHxfIxorvxpeP_csxAZgsLNC/view?usp=sharing
- It's the same as https://pastebin.com/5BwzgFz6. There are examples at the top of the file showing what it's for. This is what I used to rename all of the files.
The script requires Python, and it needs to be run in the command prompt or in Powershell. I'm not sure if you'll need the script or if you have your own tools to work with large batches of files.
>>
>>36419224
Downloading now, will take a look when it's done.

>I'm not sure if you've had more thoughts on sorting these over the last few days.
Not really, was mostly saving those thoughts for when I actually get to work on it and better understand what I'm working with and what needs to be done. The four basic categories you describe sound simple enough.
>>
File: pizza.png (545 KB, 925x1024)
545 KB
545 KB PNG
Caddy's Peetzer Parlor
https://u.smutty.horse/lzctgnppcpe.wav
>>
File: LinkSnicker.jpg (18 KB, 256x353)
18 KB
18 KB JPG
>>36419501
Hahaha
>>
>>36416398
If I'm understanding this correctly, CLIP is what gets trained on text/image data, correct? So before actually using that notebook I'd have to train a CLIP model on pony image/text data? I don't mind doing the repetitive work but I'm not exactly knowledgeable enough to set something like this up myself.
>>
File: 1483388547542.png (211 KB, 920x690)
211 KB
211 KB PNG
>>36419501
blasphemer, the real cadence loves pineapple on peetzer
>>
File: Twifags.png (371 KB, 900x630)
371 KB
371 KB PNG
>>36418035
should post this in the pony transformation thread.
>>
>>36418690
Cringe chrissy is cringe
>>
>>36418853
Oh I see. That looks like a pain in the ass, but at least we have the ability.
>>
>>36419831
Based bait-poster
>>
>>36393274
What does that mean?
>>
>>36403853
Bots have shown more self-awareness than Tumblrites and Redditors in the past.
>>
Bumping with some more hilarious angry chryssi I came across randomly. I'm glad others are figuring out how salty she can be.
https://files.catbox.moe/42ck1z.mp4
posting with catbox because smutty horse is down for some reason
>>
>>36421679
I wish more people would, since I can actually embed catbox on this browser unlike smuttyhorse
This is good.
>YOUR MOM IS GAY!
my sides
>>
Jamaican Chrysalis https://vocaroo.com/1duVMvZ9wkt2
>>
>>36421753
kek
>>
>>36421753
>Jamaican
https://youtu.be/IifWa_VnjFA?t=18
damn, i did see other anons try force models speak with accents using arphabet and edit every word individually, but I wonder if there is an easier way to just get the models speak accents with?
>>
>>36421753
nice
>>
>>36414309
awwweeee
>>
>>36421771
I've been messing with accents since the beginning, Pinkie was the best for it way back when though.

You just need to get phonetics right and everything else falls into place. Was able to get Pinkie to beat box a bit too haha.
>>
>>36406416
Starlight's the poster pony of Reddit and the enemy of the MLP fandom as a whole.
>>
File: 1111543.png (773 KB, 1011x1011)
773 KB
773 KB PNG
>>36390027
https://vocaroo.com/1f8GDnLqCbFL
>>
https://vocaroo.com/1g9qDBvLxhi0

I come from the /co/ thread with an offering
>>
File: Spoiler Image (110 KB, 1091x767)
110 KB
110 KB JPG
>>36390027
During these our most desperate times we shall praise the future artificial intellifence overlords and request their speedy arrival for censorship and abuse of information is at hand. More than usual, that is. I mean, at this point is anybody surprised?
Pray for the free flow of information and our collective enlightenment.
IFOWONAIOL
>>
I need more books to feed to my basement friend.
>>
>>36422291
nothing has been more depressing than watching Silicon Valley and having the concept of a completely open internet dangled in our faces then yanked away
>>
>>36422339
I hated that absurd ending to Silicon Valley too, so you're not alone there.
>>
>>36422249
Thanks for the offering
Maybe not as in character as >>36421679 but considering you are an offboarder, it's nice.
Have a (you) dear frien.
>>
>>36422368
I mean, I can't imagine how ELSE they could have wrapped that storyline up. It just hurt is all. Bad enough it's fictional, you gotta take it away from the fantasy even?
also I was trying to keep that vague in case anyone hadnt seen it
>>
so if I want to transcribe audio for 15 ai do I have to put the path to the audio files with a slash / or a backslash \
>>
>>36423314
Check literally any file path on your computer, or even the link at the top of the screen, and you will know the answer.
>>
>>36423326
anon, links use / while paths use \
>>
>>36423365
...but in here we use these /)
>>
>>36423382
(\
^:)
>>
>>36423382
damn straight
/'/) where my brovines at?
>>
>>36423326
the contribute example uses slashes while my computer uses blackslashes. it's confusing to me because I don't know which one to use
>>
>>36423477
they should both work. forward slash is generally preferred but it doesn't matter.
>>
>>36423382
>>36423396
>>36423444
https://voca.ro/13H98K17CcUJ
>>
>>36423382
how do you do my fellow brownies (\
>>
>>36423314
>>36423326
>>36423365
Windows accepts both / and \ for paths. Colab runs on Linux, which only accepts / . I'd imagine 15 trains on Linux as well.
>>
File: 1602129276712.png (317 KB, 450x375)
317 KB
317 KB PNG
>>36390027
is there already a plan for this >>36422460 the last PPP /mlp/ con segment was great wonder what are the talking points for this one
>>
>>36423821
15 should come give a talk or a lecture, that would be fun to listen to.
But done entirely in Twilight's voice.
>>
>>36423821
It doesn't feel like a year. What would we talk about this time? More of the same? A lot of the advancements are more on the secretive side and I think the animation AI is still a bit early yet.
>>
File: 1596245631093.png (125 KB, 276x183)
125 KB
125 KB PNG
>>36423839
>It doesn't feel like a year
this project turns 2 in april
>>
>>36423821
>>36423839
Could talk about the content side of it more this time. Y'know, potential applications for content, what's already being done, etc. Idk I'm just spitballing.
>>
>>36423871
could talk about the animation side that's been going on i know im intrested for it
>>
>>36423871
>>36423884
Plenty of time to figure out ideas. Maybe eventually get a planning doc up, see what Clipper/Synthbot/Cookie want to do.
>>
>>36423835
This would be pretty cool. It would be like Twilight herself explaining how her voice works, especially since the way they talk seem pretty alike.
>>
>>36423780
mind fucking blown. accepting either for paths sounds like asking for trouble to me
>>36423844
yeah but the last con was like 3 months ago.. right?
>>
>>36423780
I definitely have to email the 15 ai creator about this thank you for your information
>>
>>36423780
So linux only accepts what windows doesn't do by default
that seems deliberate
>>
File: 1601255056380.png (77 KB, 768x305)
77 KB
77 KB PNG
>>36424061
>So linux only accepts what windows doesn't do by default, that seems deliberate
>>
>>36424061
It's a legacy thing.
>>
>>36419831
>>36418454
>>
>>36421679
You should post this on Derpibooru (though they don't take MP4s, try WEBM).
>>
File: dt081213.gif (108 KB, 900x280)
108 KB
108 KB GIF
>>36424371
>You should post this on Derpibooru
>>
File: 1587457815260.jpg (103 KB, 820x640)
103 KB
103 KB JPG
>>36424371
>on Derpibooru
>>
taking a short break from transcribing Mister Roger, got about 20% overall done, however I feel like I should share this bit with anons here.
Keep creating you crazy, lovely people.

https://vocaroo.com/1gV6a4aLzLGN
>>
>>36424410
>>36424420
As much as you don't like it, derpi is still the de-facto content archival site. As long as the content isn't particularly controversial, there should be no problem.
>>
>>36424974
>derpi is still the de-facto content archival site
It can't be, no self respecting archive deletes its contents. Ever. Also, they themselves said that they aren't an archive and that they have nothing against "curating" the site.
Twibooru and Ponerpics are archive sites.
>>
>>36424974
Do you want me to post that thread where derpibooru mods revealed an anon's site history just for the hell of it? I'm not touching that site ever again with a 10 foot pole.
>>
>>36424974
>derpi is still the de-facto content archival site
but its not tho
>>
>>36424986
Then >>36421679 should be archived on Twibooru or Ponerpics.
>>
>>36425009
post it anon e mouse
>>
>>36425775
https://desuarchive.org/mlp/thread/36182559/
>>
File: 1590505036680.png (301 KB, 813x449)
301 KB
301 KB PNG
>>36425770
Looks like someone beat you to the punch and just uploaded it on all three
>>
>>36425854
This is why the golden rule of including "Nigger" in all of your works is important because it's like the bittering agent in Anti-Freeze.
>>
File: 1605649872588.jpg (104 KB, 900x900)
104 KB
104 KB JPG
>>36423835
>15 should come give a talk or a lecture, that would be fun to listen to.
no way he would...
>But done entirely in Twilight's voice.
yeah i can see that
>>
>>36423821
I've just posted about it in the con thread >>36426007. I've not had much time to think about this recently so throw whatever ideas you have at me over there.
>>
>>36424061
are you retarded?
>>
>>36426014
honestly who isn't on this board. /mlp/ doesn't really attract the greatest minds known to man.
>>
File: 1604336454815.jpg (91 KB, 540x540)
91 KB
91 KB JPG
>>36426027
were all retarded here but were also autistic and god knows were more organized than /g/ or /co/
>>
>>36426046
get rid of the barbies and give us mods that care, and this board could become great again.
>>
>>36426052
>mods that care
my sides bro
>>
>>36426071
>and this board could become great again.
yeah the whole thing was really a pipe dream
>>
>>36423835
>>36425992
Yes
>>
>>36424218
Unix has been around that long? Why do they act like they're the new hotness and other things are the old lameness?
>>
>>36419224
What do you think about making a panel on /mlp/con about this or doing it along the PPP one?
>>
>>36424770
hot diggity daffodil
>>36425782
jesus
>>
>>36426358
>>36426007
>>
>>36426370
I was implying there should be a section for Synthbot's animation stuff in the panel.
>>
>>36423177
I was trying to be funny more than be accurate. Appreciate it
>>
File: Two Twilights.png (514 KB, 3627x2040)
514 KB
514 KB PNG
>>36426356
Do you know what Unix Time is? It's a string of numbers that count every second that has passed since January 1st 1970, (Dec 31 1969 if you live in the west coast) which was the Unix Epoch. 4chan saves all images in this format. Yes, it's not just random numbers when you save files from this site.

Go ahead, save my image, remove the last two numbers from the filename (which are 4chan specific numbers so filenames aren't duplicated), and put that number into a unix time converter https://www.unixtimestamp.com . It'll show the exact time and date that image was uploaded.
Unix has been around for a long time.
>>
>>36426447
>all these steps telling me to do shit i don't care about
that's okay i'll just take your word for it schizo
>>
File: The_More_You_Know.jpg (29 KB, 800x450)
29 KB
29 KB JPG
>>36426447
Actually the last three numbers, which are randomly generated so images uploaded in the same second have different filenames.
And that's how the image swap bug works. If two images are posted in the same second on the same board, there's a 1 in 1000 chance that they'll end up with the same filename. There can only be one thumbnail called 1611004513128s.jpg so the two images across different threads will have the same thumbnail, but the one with the wrong one will open to the image it's supposed to be if they have different file extensions (like one jpg and one png) since the two files can then coexist.
>>
>>36426485
>Asking a question
>responding with meme words when you get a good answer

The brainlet levels are off the fucking charts.
>>
>>36426415
I'm in. If we can carve out a section for it, I do intend to talk about animation there. We mentioned it in last year's panel, so I think people will expect it.
>>
>>36426447
interesting. but weird.
>>
>>36426674
It was a different person ya goon.
>>
Someone posted the 15.ai link on /co/
Thread hit the bump limit before I could share this.
So I figure I'd just share it here.
This is the worst thing I've ever made.

https://vocaroo.com/1h4jhDUE2SH4
>>
>>36426447
Not just Unix either. A lot of the underlying tech in personal computers and the internet has been around for a long time now. TCP/IP is 1970s technology. Bell Labs created a ton of stuff around the same time: the C programming language, Unix, Optical fiber, CCDs the list goes on.
People think of the PC revolution as Apple and Microsoft thing in the 80s and 90s, but in large part they were just adapting already-existing tech to a more consumer-friendly format.
>>
>>36427085
Noice.
>>
>>36427324
>making gordon ramsey a ponycorn
>not a ram
come on man
>>
>>36426356
nigga you retarded? Unix is the operating system that most operating systems are based off of. Mac OS X, iOS, Android, linux, BSD, etc... Pretty much everything except windows
>>
>>36426632
Oh so it's the last 3 numbers. I had some filenames that only had 2. Guess a third one is there sometimes. I knew about the extra numbers at the end because it makes sense that two files can't have the same name.
>>36427115
>A lot of the underlying tech in personal computers and the internet has been around for a long time now.
Same way x86 is still in use today, although it's more for backwards compatibility. Every time Intel tries to kill it (Itanium for example) it just gets stronger. A 42 year old instruction set is still in use today. I can run MS DOS just like the original PC could 40 years later.
Apple has recently moved away from Intel and has introduced ARM based desktop processors for their new Macs. Something about making their computers more compatible with iPhones.
>>
>>36427517
It's because they generate a random number between 1-999 to prevent duplicates
>>
>>36427544
I always thought since it's three numbers it's milliseconds.

>>36427517
Apple making ARM desktop chips is definitely a very interesting development, Microsoft tries to make an ARM Windows every few years (currently at it again I think), but have failed every time so far because the x86 emulation just isn't there. Apple has it far easier there due to the much more controlled system and actually making the hardware themselves. It might kick something off, but it also might not considering they also used PowerPC for a long time.
RISC-V when?
>>
>>36427440
Had no idea! Learning a lot today. My knowledge of computer history is really vague prior to the IBM era.
Like I said, the haughty linux users act like they're a new, innovative thing, not an old thing.
>>
>>36390027
Has anybody tried to mess around with automatic speech alignment with the AI voices?
Could you use your own voice to better time and sync up the generated voices?
Not sure how it works, just had the idea.
>>
>>36428659
Someone, yes! Two threads ago he teased us with it, we haven't heard from him since.
>>
>>36428659
People have played with it. I know one person recently posted an example of a tool they developed to do just that (Here: https://desuarchive.org/mlp/thread/36358165/#q36369386), but I believe all they did was post the example and then disappear.
>>
>>36428710
After some google searching, looks like adobe audition and VocALign has features to do it, though I doubt they have robustness to go from human->ai.
How difficult could it be to make own algorithm to do it though. I would assume it would be 'just' extracting features and minimizing the differences of those features by controlling tempo and cutting out or adding silent pieces.
P.S. no idea how audio works but just guessing

>all they did was post the example and then disappear.
faggot
>>
>>36428659
Due to the way FastSpeech2 works you can probably insert per-phoneme durations and have it figure it out. Combine that with an ASR model/forced aligner trained on the same phoneme set and you can probably feed said durations and phonemes to the model, bypassing the duration predictor.
Talking purely theoretically. I haven't tried it.
>>
>>36424770
Holy shit. Godspeed, anon.
>>
>>36429307
I honestly don't get that phrase. How fast is that?
>>
>>36429845
retard
>>
>>36426447
What's with pegasus Twilight?
>>
>>36392238
Thanks. If you're not working on it (or if you are, and want to split the work), I can do some after I'm done with this other set.
>>
File: Two Twilights - Alicorns.png (519 KB, 3627x2040)
519 KB
519 KB PNG
>>36429945
I was rendering a pegasus Twilight to see what it would look like if she had "normal" pony wings instead of her Alicorn ones. Here's one with the horn. You can see how Alicorn Twilight is longer than the Unicorn Twilight I just put wings on in the left, and she has more of a neck.
>>
>>36430051
pegacorn master race?
>>
>>36430064
Yes. Here is a Pegasus sized Alicorn Twilight with open wings.
>>
>>36430064
>>36430138
Half open
>>
>>36430097
>Alicorn Twilight
Put it in the garbage.
>>
Post
>>
>>36430934
didn't work try again
>>
>>36416398
>Maybe someone here can train it on pony images
How? I went into all the related documentation I could find but can't really find anything resembling a way to train a model on custom images.
>>
I think I can use this for my VN
>>
>>36423909
More 15.ai content in this thread
>>
>>36431272
Use what?
>>
>>36431792
the voice generator for the lines
>>
>>36430141
>>36430138
>>36430051
it's so weird how most of her actually gets smaller/thinner because they didn't want to make her hugely taller
>>
https://www.youtube.com/watch?v=r0FLyW86owo
>>
new thread when
>>
>>36432078
We're still at page 5 anon
>>
File: 1531811150981.jpg (41 KB, 500x500)
41 KB
41 KB JPG
>>36431734
I told her to learn to sing first

https://voca.ro/104KNNQwm2aa

>>36432239
Well, not anymore
>>
File: bad fur day.png (165 KB, 836x955)
165 KB
165 KB PNG
Oh shit we hit bump limit
Any backer here?
It's late and I won't be able to do it
>>
NEW THREAD
>>36432529



Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.