Cool and/or Useful AI Tools

[Colon capital V]

With all the talk about Stable Diffusion and Machine Learning art finally giving us a peak into the future, I thought it'd be good to post any other AI tools that are fun to play with or might have some legitimate use for any of us.

I'll start off with one that I've had some use for a long while:
Waifu2x

Image upscaler/denoiser originally used to clean up anime art but can be used on any low-res photo to (hopefully) look better.
Here's the one for manual installation: https://github.com/nagadomi/waifu2x

And if you're on windows and want a gui instead of fiddling with the console, there's waifu2x-caffe: https://github.com/lltcggie/waifu2x-caffe

Spleeter

Music tool that splits up stems of a track. Can be useful if you're into music making, trying to get certain songs from tv shows or whatever else.

GitHub - deezer/spleeter: Deezer source separation library including pretrained models.

Deezer source separation library including pretrained models. - GitHub - deezer/spleeter: Deezer source separation library including pretrained models.

github.com

Again if you don't wanna bother setting it up, there is a web varient of it but I can't tell you if it's better or worse than where it's originally forked from: https://splitter.ai/

They have a "2-Stem PRO" subscription which claims to do better at splitting vocals from the track, but I'd recommend Ultimate Vocal Remover since it's free and I'd imagine does a much better job than what Spleeter or Splitter.ai can do.

MuseNet
Created by OpenAI, this site let's you put in any midis you'd like (given that it's short enough) and it'll do its best to continue the song based on the style you choose for it. Pretty fun to mess around with and see what results you can get: https://openai.com/blog/musenet/#fn2
Ex:

Jukebox
Also created by OpenAI, if you can set this up this lets you input any music tracks and have it try to continue like MuseNet and the results can vary wildly.

Jukebox

We’re introducing Jukebox, a neural net that generates music, including rudimentary singing, as raw audio in a variety of genres and artist styles. We’re releasing the model weights and code, along with a tool to explore the generated samples.

openai.com

Tutorial on how to set it up + video:

Google Colaboratory

colab.research.google.com

OpenAI Jukebox: One-Click notebook - Quick Guide

One-Click notebook:https://colab.research.google.com/drive/1sJda9v46gNzBc7m59MP5zn63AWc-axCY?usp=sharingLists of available artists and genres (5b_lyrics uses...

www.youtube.com

Feel free to suggest other AI tools/programs that are fun to mess around with.

 

[Stephanie Bustcakes]

Uberduck is a text-to-speech ai that has voice data sets of various people and characters (of varying quality). This means you can make your favorite personalities say copypastas and gamer words and stuff.

 

[Epic Fail Man]

15.ai was pretty fun.

 

[Baraadmirer]

15.ai was pretty fun.

View attachment 3749265
View attachment 3749268
View attachment 3749270
View attachment 3749272
View attachment 3749274

I'm pretty sure that the iconic Rise Kujikawa copypaste was done by 15.ai.

 

[Slav Power]

If we're speaking upscaling, it's worth mentioning ESRGAN, which can be used to upscale images, textures, or even remove various specific artifacts, such as JPEG compression, DXT1 compression, resizing artifacts and more.

It's best to visit this wiki for more info: https://upscale.wiki/wiki/Main_Page
Generally what you'll want to use is Cupscale: https://github.com/n00mkrad/cupscale
And also you'll want to get a nice database of models from here to use: https://upscale.wiki/wiki/Model_Database

 

[Baraadmirer]

If we're speaking upscaling, it's worth mentioning ESRGAN, which can be used to upscale images, textures, or even remove various specific artifacts, such as JPEG compression, DXT1 compression, resizing artifacts and more.

It's best to visit this wiki for more info: https://upscale.wiki/wiki/Main_Page
Generally what you'll want to use is Cupscale: https://github.com/n00mkrad/cupscale
And also you'll want to get a nice database of models from here to use: https://upscale.wiki/wiki/Model_Database

Does this have the same effect as sharpening images? I'm looking at Neural Enhance or what Vance AI can do; anyone know how good it is or anything similar?

 

[Colon capital V]

Forgot to give a special mention to Whisper, also by OpenAI, that subtitles videos of any language and was apparently trained on over 680,000 hours of multilangual speech across the web.

 

[Slav Power]

Does this have the same effect as sharpening images? I'm looking at Neural Enhance or what Vance AI can do; anyone know how good it is or anything similar?

Depending on the model you use, it can sharpen the image. For example the Universal Upscaler model has two extra models for 4x upscaling with extra sharpening. What you get out of ESRGAN depends on what models you use, in which order, and what the initial size of the image you're upscaling is.

 

[Colon capital V]

Bumping this thread again just so I can post these for later:
Character.ai - Fun little website for you to dick around and role play with artificial characters in whatever scenario you go with. (Though it is censored to a certain extent so that you can't get too horny or "based" with the characters or else you run the risk of their messages being deleted outright.) Requires a signup for you to continue messaging but nothing like a temporary email won't do.

Adobe Podcast Enhance - This tool by Adobe will try to enhance any recording you throw at it to the best of its ability. Definitely useful for any old recordings you'd like to "restore" using it, although it does require you to have an adobe account but once again a temporary email will do just fine.

 

[evarus]

Bumping this thread again just so I can post these for later:
Character.ai - Fun little website for you to dick around and role play with artificial characters in whatever scenario you go with. (Though it is censored to a certain extent so that you can't get too horny or "based" with the characters or else you run the risk of their messages being deleted outright.) Requires a signup for you to continue messaging but nothing like a temporary email won't do.

Very funny site