Old Instructions for C
- Build it yourself
- Go in the root directory of the repo and run either of these commands
make
- OR
gcc -Wall src/main.c -o bin/main -lX11 -lXtst
- Then run the binary
./bin/main <song prompt>
Simple Note
- This script utilizes the Firefox browser and will automatically download the binary and required files into cache (Linux: /home//.cache)
- I am assuming that you are already logged into https://suno.com or https://riffusion.com and hence we need to maintain the cookies in the puppeteer instance of firefox as well
- Since puppeteer always opens a new temporary profile, we have to pass the User data directory that firefox generates into
src/main.js - I have provided the sample directory name for your reference.
- Do make sure that you change that to your directory before running the program
- Furthermore, In order to prevent file locking, close all other firefox instances
- Add Puppeteer and download suitable firefox binary
make
- OR
yarn add puppeteer
npx puppeteer browsers install firefox
-
Put your user data directory for Firefox in
src/main.jsat line 59:userDataDir: <path>- For Windows:
userDataDir: "C:\Users\<username>\AppData\Roaming\Mozilla\Firefox\Profiles\*.default-release" - For Linux:
userDataDir: "/home/<username>/.mozilla/firefox/*.default-release" - If you are not able to find your directory, just type
about:profilesin Firefox's search bar and see the path given as Root Directory
- For Windows:
-
Ensure you are logged in to the respective AI song generator in Firefox
-
Ensure that you have unchecked this option in firefox's settings
Always ask you where to save filesso that firefox doesn't prompt you everytime the script downloads the song -
Close all other instances of firefox
npm(package name: nodejs)yarnmake(optional)
node src/main.js </path/to/prompts>
I am being forced to use python (which is not even a real programming language) since it provides a good abstraction layer over HID control /s
- Main Objective: Automating extracting data from websites
- Note that for this automation, you need a dedicated system since I will be using this program on websites which have a paid API
- Most websites have a limit on the requests we can send per day per account so I will be using https://temp-mail.org for making new accounts
- They have a free API called RapidAPI which lets us send 100 reqs per day which is more than enough for our use case
- So, I will be using temp-mail to generate new accounts once limit has been reached on the current account
- For now, I am targeting www.udio.com
- udio has added a captcha in order to prevent automation and guess what? its working. I am not able to bypass it
- Hence, I am changing my target from udio to suno ai for now
- completed the script for automating downloading ai generated songs from suno ai
- optionally, you could install
dunston your system to see the timer countdowns
NOTE: disable "Always ask you where to save files" in your browser so that the browser wont prompt you to provide download path every time it downloads the songs
- Difficulties faced with using C with XLib for browser automation
- hard-coded values for time countdowns are either way too long for good performing machines (time wastage) or way too short (completely breaks down the script) for low performing machines
- On some systems, I just cant seem to bring focus on the buttons which needs focus due to unknown reasons
- The script is very performant but not at all flexible to work on "all" systems
- I might be switching over to the dreadful language of all time, JavaScript (with Puppeteer framework) because of its great abstraction over web utilities which aid immensely for web automation
- While trying to figure out how to detect when generated audio is ready for download, I have noticed that https://suno.com is now asking for captcha when trying to generate the song
- So, suno.com is also now out of the options along with udio.com
- A major advancement made was shifting from C to JS (with puppeteer) since it makes the program very flexible and easy for web automation
- I was able to do the following with the Puppeteer JS framework with ease...
- Open up firefox/chrome/chromium browser
- Wait till it has properly opened
- Go to a specific website (suno.com) in our case and again, wait only the required time to load the webpage
- Detect the textbox and enter text into it and then click the submit button without the need to hardcode any values
- Detect when the loader spin document element goes out of the DOM so that we can download the songs
- Detect any Captcha popups
- But the problem I faced with this method was that even though I was initially able to generate the songs, after some tries, the website might have put a stronger bot tracker on my IP or something because of which I started getting a captcha every single time
- To conquer this, I tried to load the Stealth plugin of the Puppeteer framework alongside adding a little bit of delay when typing each character and clicking the generate button to try to deceive the bot tracker but even that did not work
- I even tried to run the script through a headless browser but that method failed as well
- However, I was able to generate the song without any captcha if I did all the steps manually
- So, I concluded that the website must be able to track that I am running a script somehow
- To try to solve this issue, I will be going back with the previous method and try to manipulate HID (keyboard and mouse) to get the final result using C
- I would have really preferred writing the script in C but that would mean I will have to use some library to manipulate HID like Xlib. But the problem with that approach is that Xlib only works for X11 display manager obviously but 2 people in our group run Windows as their primary Operating system and since Windows does not run X11 as its display manager, my script won't work for them
- So unfortunately, I will have to move to a language which I don't even consider a programming language, Python
demo_python.mp4
- I am able to hit the download button and go back to the text prompt in order to generate more songs but the actual download time is quite high (about ~100s)
- https://suno.com was giving captcha with puppeteer and hardcoding mouse positions with python (and pyautogui) was fine but would not work on everyone's system out of the box without any configuration
- Hence, we shifted to https://riffusion.com which currently has no limits and does not give a captcha even with Puppeteer
- And as always, I am writing the development log after the event had occurred, so, the JS script is ready
- The JS script is working as intended and as required, it is picking up the prompts from a separate file which contains newline separated prompts
- Problem: riffusion sometimes blocks the request due to high traffic and since my script does not account for that, it continues to download the top 2 prompts (which in this case would already have been downloaded since no new songs could be generated) and proceeds to the next prompt
- To solve this issue, I must keep track of the songs downloaded and retry generating the prompt until the song is downloaded and until we reach a threshold amount of tries (to prevent infinite loop of trying to generate from prompts which cant generate a song)