Amazing job! I have some ideas for things I'd like to make by adapting your code, but I can't with no copyright info attached to the code. I would suggest the timeless GNU Public License: https://www.gnu.org/licenses/gpl-3.0.en.html

Traceback (most recent call last):
File "touhou_bad_apple_v4.0.py", line 1, in
import cv2
ImportError: No module named cv2

I never used Python but I really wanted to try this. I followed all the steps.

\Desktop\bad-apple-master\touhou_bad_apple_v3.py", line 7, in
from PIL import Image
ModuleNotFoundError: No module named 'PIL'

block=False is not supported in Linux, you may need to add an thread to execute playsound():

import threading
...

threading.Thread(target=playsound, args=("tng_bad-apple.mp3",)).start()

At the end, theres that. :I
it works perfectly until it reaches the end
Python 3.9 has been used.
Afaik Python 3.10 has some issues cause this hasn't been optimized for Python 3.10 or newer.

Traceback (most recent call last):
  File "/home/hax/GitHub/bad-apple/touhou_bad_apple_v4.5.py", line 181, in <module>
    main()
  File "/home/hax/GitHub/bad-apple/touhou_bad_apple_v4.5.py", line 171, in main
    play_video(total_frames=total_frames)
  File "/home/hax/GitHub/bad-apple/touhou_bad_apple_v4.5.py", line 37, in play_video
    sys.stdout.write("\r" + ASCII_LIST[frame_number])
IndexError: list index out of range 

Forked your repo, and had a look, also had a look about the suggestions made by the other user, but honestly, cv2 and the way you're currently doing the ascii conversion, is over my head slightly. If I properly sat down and read it, I might get somewhere, but anyways.
I've tweaked your script to also use threading on the frames, so it should be faster. Also, completely removed the need to save to a file, twice, by just directly passing the cv2 result to PIL, and then outputting the frames, though I do like the idea for keeping the frames, which is why I'm also looking into perhaps RLE compressing the frames into a single file, instead of having a whole folder of 7kb ASCIIs.
I'm not really sure if you wanted the tweaks I made in your version, so I made it as a separate file on my repo and put this as a issue, instead of a PR (though if comments were a thing, I'd put it there), if you just want to take a look. Just as someone who vaguely knows Python, this was a nice project to at least be able to work on.

Oh, and progress bar still works!

Hey :) Its been a while since my last ticket :)
I have a new one for yaa. Even when this project was a weekend project. im still amazed about your progress..

It seems to be that there is something wrong with the audio format...

see log:

python3 touhou_bad_apple_v4.5.py

Select option:

1. Play
2. Exit
   ==============================================================
   Your option: 1
   Please enter the video file name (file must be in root!): BadApple.mp4
   MoviePy - Writing audio in audio.mp3
   MoviePy - Done.
   Beginning ASCII generation...
   Progress: [####################### ] 99% Frame 6570 of 6571 framesASCII generation completed! ASCII generation time: 28.546977519989014Traceback (most recent call last):
   File "touhou_bad_apple_v4.5.py", line 181, in
   main()
   File "touhou_bad_apple_v4.5.py", line 170, in main
   play_audio('audio.mp3')
   File "touhou_bad_apple_v4.5.py", line 24, in play_audio
   pygame.mixer.music.load(path)
   pygame.error: Unrecognized audio format

Hi, I saw the code on touhou_bad_apple_v2.py and I though of some possible speed improvements right off the bat.

First is ASCII generation. As you're doing right now, the best way to improve processing speed is by creating a Pool and allocating each frame to a worker until the frame extraction is over. The disadvantage is that you lose the current progress bar you have implemented, and instead will need something more complex like an N lines progress bar, one for each worker in the pool, or a signaling method between threads, like a semaphore, and is just too much work for what the current project is aiming for. You can also improve the video quality by adding more characters to the list, more specifically a space(all black) " " and a block character(all white) "\u2588", and adapting the padding calculation on line 86 to include the new characters.

Another improvement, and probably the most important one, is saving the ASCII frames to a single file containing all frames, instead of multiple files. The reasoning is that opening and closing files takes time, which would be better spent synchronizing the frames in the terminal.

Checking files also takes time. As of right now, you are manually checking every single .txt file for all the frames and checking if they exist. This can be done much faster if you use instead os.listdir() and filter the results containing the string .txt, then count the results(or a single file if you use the approach above). Also, please always use os.path.join() instead of manually using /. Reasoning is that the os.path library provides cross-platform separators and better handling of absolute paths, which is much less error prone.

I could make a fork and do those changes, but I think it would be very useful if you did them yourself, since you can learn new skills and functions with them. Tell me what you think about my suggestions, then close this issue later.

Edit: my mistake, I just noticed you changed to a white background in the video. Correct characters are a space(all white) " " and a block character(all black) "\u2588"