YouBit allows you to host any type of file on YouTube.
It does this by creating a video where every pixel represents one or more bits of the original file. When downloaded from YouTube, this video can be decoded back into the original file. Here's an example of such a video. YouTube imposes no limits on the number of videos you can upload.
This is not a novel idea and has been explored by other projects such as YouTubeDrive and fvid. However, these projects left alot of good ideas on the table. YouBit adds a bunch of features, ideas and options to the base concept, while making a reasonable attempt at being performant.
pip install youbit
NOTE: If a wheel is not available for your platform, a C compiler needs to be installed prior to installation.
Encode and upload a file:
python -m youbit upload C:/myfile.txt firefox
NOTE: If you want YouBit to upload videos for you, chrome needs to be installed.
The 'firefox' argument denotes which browser YouBit should extract cookies from to authenticate itself with YouTube.
In this example, the currently logged-in account in firefox will be used. Can be any of ('firefox', 'chrome', 'edge', 'brave', 'opera', 'chromium').
Make sure this account has gone to studio.youtube.com before using it with YouBit.
When going there for the first time, some popups might appear to ask for things like a channel name. YouBit will not traverse these for you.
Download and decode a YouBit video:
python -m youbit download https://youtu.be/dQw4w9WgXcQ
Output is always saved in the current working directory unless specified otherwise.
Decode a local video:
python -m youbit decode C:/myvideo.mp4
NOTE: This video needs to have been downloaded from YouTube. You cannot encode a file, and then immeadiately decode just to test it out.
Encode a file without uploading:
python -m youbit encode C:/myfile.txt
Alternatively, the Python API can be used directly:
from youbit import Encoder
from youbit.settings import Settings, Browser
settings = Settings(browser=Browser.CHROME)
encoder = Encoder('C:/myfile.txt', settings)
url = encoder.encode_and_upload()from youbit import download_and_decode
filepath = download_and_decode('https://youtu.be/dQw4w9WgXcQ', 'C:/mydir/')Encode without upload:
from youbit import Encoder
encoder = Encoder('C:/myfile.txt')
filepath = encoder.encode_local('C:/mydirectory/')Changing encoder settings:
from youbit import Encoder
from youbit.settings import Settings, Resolution, BitsPerPixel
settings = Settings() # sensible defaults if left untouched
settings.resolution = Resolution.QHD
settings.bits_per_pixel = BitsPerPixel.TWO
settings.ecc_symbols = 69
settings.constant_rate_factor = 20
settings.null_frames = True
encoder = Encoder('C:/myfile.txt', my_settings)Decoding a local file a little trickier. You must know the settings that were used during the encoding process in order to succesfully decode it. You do this by passing a Metadata object to the decoding function. Your YouBit videos should have a base64 encoded string as their description. This string can be used to instantiate the correct associated Metadata object.
from youbit import decode_local, Metadata
metadata = Metadata.create_from_base64(VIDEO_DESCRIPTION_STRING)
filepath = decode_local('decode_me.mp4', 'C:/documents/', metadata)- Does this mean infinite, free cloud storage?!
- Why no colors?
- What is 'Bits Per Pixel'?
- Why a framerate of 1?
- Why not you use the YouTube API for uploads?
- After uploading, how long do I have to wait to download A YouBit video again?
- Why can I not use resolution x?
- How large can my file be?
- Why not upload lossless videos?
- What settings should I use?
No.
- It's slower: encoding and decoding takes time. The files uploaded to YouTube are much larger than the original. YouTube needs to process the video.
- You can't trust it: If YouTube changes some things tomorrow, there's a chance your video can no longer be decoded.
It's just a very fun concept to explore :)
Because chroma subsampling will compress away color information with extreme prejudice. So instead we save all our information in the luminance channel only. This results in greyscale videos, and works much better. It coincidentally makes the encoding and decoding process less complex as well.
As you might have guessed, BitsPerPixel is a settings that dictates how many bits of information will be saved in a single pixel of video. A higher BPP allows for a higher information density - a smaller output video in comparison to the original file. However, it also introduces more corrupt pixels.
A BPP of 1 means each pixel only has 2 states, 1 and 0, on and off, white and black. This means our greyscale pixels have a value of either 255 (white) or 0 (black). During decoding, YouBit treats anything 128 or more as a 1, and everything below 128 as a 0. This means YouTube's compression needs to change a pixel's value by at least 127 for it to become corrupt.
Now consider a BPP of 2. Two bits have 4 possible states (00,01,10,11). So to represent 2 bits, our pixels need to have 4 possible states as well. Something like (0,85,170,255). The distance between these is now smaller: a change of only 43 is now required to corrupt the pixel. Our video will be half the size, but easier to corrupt when YouTube re-encodes it during upload.
The default settings use a BPP of 1. A BPP of 2 is possible, and 3 is experimental.
I do not know exactly how YouTube decides on the bitrate to allocate to a stream, but it seems to rougly follow their recommended video bitrates. All else equal, a video with a framerate of 1 will get the same bitrate as a video with a framerate of 30. See where I'm going with this? More effective bandwith per frame, less compression.
Secondarily, using a framerate of 1 during encoding allows us to read only keyframes during the decoding process. This is very important. Testing showed a massive delta in corruption between keyframes and B- or P-frames. Many keyframes would be completely void of any errors, while some B-frames at the end of a GOP would be almost entirely unusable.
If we use a framerate of 1, YouTube will re-encode it as a video with a framerate of 6. This seems to be the minimum on YouTube. After analyzing the (open) GOP structure of these 6fps videos, it became apparent that just skipping any non-keyframes during the decoding process is not enough. We would see duplicate keyframes scattered around. Fortunately, these duplicate keyframes are predicatable. YouBit discards what it knows to be duplicate keyframes during the decoding process.
This does mean that YouBit videos that did not go through YouTube, cannot be decoded.
There are 2 reasons. For one, unverified API projects can only upload private videos. These videos are locked to being private and this cannot be changed. This means YouBit links would not be able to be shared between users. (And no, getting this project verified by Google is not an option for obvious reasons).
Secondly, The YouTube Data API v3 works with a quota system: all interactions with the API have an associated cost. Uploading a video costs a staggering 1600 points, out of 10,000 points that are replenished daily. This would limit the user to a measly 6 uploads per day.
Instead YouBit extracts cookies from the browser of choice, to authenticate a Selenium headless browser instance where the upload process is automated. This is very hacky, adds alot of overhead and is very sensitive to changes to YouTube's DOM, but it is the best we've got.
This is tricky, since it can take a long while for YouTube to fully finish processing a video. If the video is unavailable because it is still processing (this means not even an SD stream is available yet), YouBit will throw an exception. If the video is technically available, but the resolution that was specified during the encoding process is not yet available, YouBit will throw an exception. If neither of the above, YouBit will download the video and attempt to decode it.
Thus, it is recommended to wait a sufficient amount of time. The highest available video bitrate (VBR) of any uploaded YouBit video can be checked most easily using the CLI:
py -m youbit test vbr https://www.youtube.com/watch?v=SLP9mbCuhJc
| Resolution (no zero-frame) | VBR settles around |
|---|---|
| 1920x1080 | 10200 |
| 2560x1440 | 19700 |
| 3890x2160 | 47800 |
| 7680x4320 | 172407 |
The decoding process might very well still work with a lower VBR, it all depends on the settings that were used.
There's no real advantage to using higher resolutions than the default of 1080p.
Technically YouBit can work with any resolution video, however, resolutions lower than 1920x1080 are not supported because the bitrate YouTube allocates to them does not always scale favourably for our use-case. It also introduces inconsistencies because YouTube may use different, less performant codecs based on the resolution.
Resolutions higher than 1080p are supported, but should generally not be used. If you wish to experiment these might be of interest to you, but otherwise they just introduce a far longer processing time on YouTube's end.
YouBit encodes your files in chunks, so you are not limited by system memory, but we are limited by YouTube's maximum video length. YouTube videos are allowed to be up to 12 hours long, or 128GB, whichever comes first. YouBit will raise exceptions during the encoding process if either of these are violated. (If the YouTube account you are using is not verified, the limit is 15 minutes instead. Be sure to verify your account.)
Ofcourse, this depends entirely on the settings selected.
To give you an idea, the default settings will stop working with files larger than 9GB.
YouBit has the option to use 'null frames'. If enabled, YouBit will interject completely black frames in between 'real' frames when generating the video. The idea is that YouTube will still allocate the same bitrate, but since the video is twice as long and all-black frames can be compressed away almost entirely, we will have twice the effective bandwidth per actual data-holding frame. In practice, this only works a little: videos with zero frames have a lower bitrate, but not half. 1080p videos seem to get a bitrate of 7000, compared to the usual 10200.
This is still a useful ~40% effective inrease in available bandwidth, leading to less errors and a potentially higher information density.
On higher resolutions however, the use of zero frames seems to be detrimental. The default settings forego using null frames.
Use at your own discretion.
Compressing the video locally (before YouTube will compress it again) might seem like a very bad idea if we want our data to remain intact. However, the difference in filesize is very big. And as soon as the encoding process is reasonably time efficient, the time it takes to upload the video to YouTube becomes by far the biggest bottleneck. If we carefully control the amount of compressing that we do locally, we can make our video alot smaller (and faster to upload) without affecting data integrity all that much.
That being said, changing the 'crf' option in YouBit allows you to control this variable. It is simply the Constant Rate Factor setting that is passed to the x264 codec.
Unless you know exactly what all the options do and how they interact, I would advise you to stick to the defaults. Not to say that the defaults settings are the best, they are just the simplest and most reliable. You can actually speed up the whole process (encode -> upload -> download -> decode) significantly by using different settings.
For example, setting 'null_frames' to True allows you to use a 'BitsPerPixel' of 2, which drastically speeds up the whole process, since the resulting video is half the size while still being able to be succesfully decoded.