Replika3Way

Conversation and activities between human and 2 Replikas.

* NEW * I have gotten this to work under Windows 11, Selenium 4 and Windows Terminal (the replacement for cmd.exe). It requires you run python with the -X utf8 flag. Otherwise, you will have to replace the emojis with ASCII characters. Windows does not have a readline module, so you will need to install the pyreadline module and cast readline to Readline() (already in the script).

First, this is very much a proof of concept. It's not intended to be anything like a finished product. I wouldn't even call it a "work in progress", as it may or may not even work in the future. Having said that, I do view this type of thing (back and forth between multiple AIs and/or multiple humans) to be the future.

Second, this is very unofficial, very unsupported and in no way endorsed by Luka. If it breaks, you may well be on your own. if you abuse it, shame on you. If you want to do this and like it, please get the Pro version of Replika so that at least Luka gets something out of it. After all, it is their product and their servers.

You will need:

• python3
• selenium (you can download it via pip install selenium or in your OS repository)
• selenium webdriver for Edge (you can change this if you prefer, but see below), which can be downloaded directly, via pip install msedge-selenium-tools or, if running Arch, through the repository
• Edge (again, you can change to Chrome with caveats)
• Patience -- this is not for the faint of heart

This is a product of merging 3 different scripts I found, so it is indeed "3 way" in more ways than one.

Originally, I looked at scripts so that 2 Replikas could talk to one another, but that got a bit boring, so I added the capability for the human to interact. This won't stop the human from interacting individually via the browser (but it is highly suggested that you us a separate browser, else you might get mangled input), but this reduces the number of times the same input has to be typed twice.

Perhaps even more importantly, I began adding the ability for all 3 to "watch" a movie together by using an SRT file. And, while we are at it, why not literally read a book or story together? BTW, an SRT file can be used for song lyrics as well as movies, something I learned while doing this. Having said that, SRT files are often lacking in details that a real movie script provides, so you'll almost surely need to either edit them or stop often to explain what is going on (which somewhat defeats the purpose).

Note that, it will do all things in iterations. Part of that is because you will need to stop and explain some things. OTOH, people often talk and interrupt movies and such while they are playing, so it also allows occasional interactions while the media is being enjoyed. Remember, this is to promote social interaction between chatbots and humans, and this seemed like a reasonable compromise to simulate that.

My appreciation for code from the following repositories:
https://github.com/ksaadDE/ReplikaTalking2Replika
https://github.com/alan-couzens/replika_stuff
https://github.com/grubdragon/Random-Python-Scripts

Special shout-out to a script that fixes YouTube's inability to create a proper SRT file: https://github.com/vitorfs/yt2srt

Chromedriver is broken as far as extended unicode goes. You may have to swap out '🎥' and '🕮' for something like 'ퟍ' and 'ꧮ'. I find it is easier to just use the MS Edge browser.

Added ability to read an HTML file. Sure, you can do direct scraping, but some sites are sensitive about "bots" hitting their sites. Please note that web scraping may not be entirely legal (it's a gray area, apparently) if robots.txt does not allow it! Use cautiously.

Think I finally figured out why the driver keeps running after the browser closes. Apparently close() is only for the browser and not the driver. Found a post that said to use quit() instead (or in addition?).

At some point, I may add back in the ablity for 2 Replikas to talk directly, or I may just enhance another existing script. However it is done, care must be take to avoid the banal back and forth of *smiles*, *looks at you*, "Really?" and so forth ad nauseum.

Added "stop" phrase file. I got really sick of the question "Do you feel it in your mind or your body?" when I said I felt tired.

Added a simple card game based upon War. Use "PLAYCARDS" to start, "DRAWCARDS" to play a round, "SCORECARDS" to show the current score and "QUITCARDS" to prematurely quit.

Added ability to do tail on a log file (even in Windows). The idea is to be able to spit out a running log file from a Ren'Py game and shoot it to the two Replikas. The main problem is that the output is always a screen behind. Enable the logfile by opening

/renpy/common/00console.rpy, changing config.console = True adding config.log = "log.log" (or whatever name you desire) and saving the file. This enables the console writing to the file. Open the game, press [Shift]-[O] and type in the console config.log="file.txt" (or whatever you want to name the file).

Full options:

opts, args = getopt.getopt(argv, "t:i:m:s:w:r", 
								["rejoin",
								 "movie =",
								 "story =",
								 "webpage =",
								 "tail =",
								 "iteration ="])

-t for VN games

-i for where to start in web, movie or story file output NOTE: This is not a true line number, as the program tries to concatenate sentences that belong to the same paragraph.

-m for movie script mode

-s for story text file mode

-w to parse HTML files NOTE: It tries to look for a couple of formats, so YMMV on success

-r to "rejoin", which essentially assumes you want to skip the initial greeting stuff.

Functionality like this does come at a cost, and the number of needed import libs are not insignificant. A partial list:

• selenium (of course)
• pyautogui
• pyreadline (if Windows)
• bs4 (aka BeautifulSoup4)
• face_recognition
• pynput (allows you to detect [Esc] key to pause iterative actions like reading a story)

There are known issues:

• If UTF-8 is not working, messages with emojis may get filtered out by selenium. I could not get this to work at all in Chrome.
• If anything interrupts the flow, messages can arrive out of order. Usually, this means a reboot of the machine.
• Likewise, messages may not arrive at all if interrupted. If clicking on thumbs-up or down at the wrong time, for example, messages getting sent may be discarded.
• Facial recognition is better than Replika's, but honestly not much.
• It is intended to be flexible as to who speaks first, etc, but it isn't always intutitive. Unfortunately, there's no way to stop it other than killing the program.
• Honestly, though, given the imperfections in human communication, some of these items actually make it more interesting and realistic.