Liberate Your YouTube Videos

If you've been following this blog, you'll know that Google unjustly shut down my YouTube channel. They've now reinstated it - but I can no longer trust them as custodians of my data.

So, here's a quick tutorial on how to download all your videos - and metadata - from YouTube.

The Official Way

Google offers a "takeout" service which will allow you to package up all your YouTube videos for export.

It creates a multi-gigabyte archive - which isn't particularly suitable for hosting elsewhere. Once the archive is created, you have to download it in 2GB chunks. The archives are only available for 7 days - so if you're on a normal speed Internet connection, you might not be able to grab everything.

If you do manage to download everything - you'll find another problem. The files are enormous because you're downloading the originals - not the web-optimised versions.

So, how can we download high-quality, low-filesize copies of the videos suitable for HTML5 use?

P-p-p-p-pick Up Some Python

We'll be using the excellent YouTube-DL - make sure you have the most recent version installed.

The Google Takeout from above contains a file called uploads.json - it has a list of every video you've uploaded and some associated metadata:

 JSON...
,{
  "contentDetails" : {
    "videoId" : "2gIM9MzfaC8"
  },
  "etag" : "\"mPrpS7Nrk6Ggi_P7VJ8-KsEOiIw/7fgl8GbhgRwxNU3aCz9jzUB_65M\"",
  "id" : "UUAEmywW2HASHP0MohSNIN0qHLpdeAxkQv",
  "kind" : "youtube#playlistItem",
  "snippet" : {
    "channelId" : "UCyC5lCspQ5sXZ9L3ZdEEF2Q",
    "channelTitle" : "Terence Eden",
    "description" : "Look at me walk to work!",
    "playlistId" : "UUyC5lCspQ5sXZ9L3ZdEEF2Q",
    "position" : 168,
    "publishedAt" : "2010-12-02T10:03:15.000Z",
    "resourceId" : {
      "kind" : "youtube#video",
      "videoId" : "2gIM9MzfaC8"
    },
    "thumbnails" : {
      "default" : {
        "height" : 90,
        "url" : "https://i.ytimg.com/vi/2gIM9MzfaC8/default.jpg",
        "width" : 120
      },
      "high" : {
        "height" : 360,
        "url" : "https://i.ytimg.com/vi/2gIM9MzfaC8/hqdefault.jpg",
        "width" : 480
      },
      "medium" : {
        "height" : 180,
        "url" : "https://i.ytimg.com/vi/2gIM9MzfaC8/mqdefault.jpg",
        "width" : 320
      }
    },
    "title" : "Walking To Work In The Snow"
  },
  "status" : {
    "privacyStatus" : "public"
  }
}, {
...

So, we want to go through that JSON, download some web-friendly versions of the media, and save them.

This Python downloads a high resolution MP4 video and AAC audio - it then mixes them together. It will also download a slightly lower resolution WEBM file. It then grabs a screenshot and any subtitles which are present. Finally, a little HTML5 snippet is written.

 Python 3from __future__ import unicode_literals
from datetime import datetime

import youtube_dl
import json
import urllib
import os

#    Read the JSON
with open('uploads.json') as data_file:
    data = json.load(data_file)

#    Itterate through the JSON
for video in data:
    videoId     = video["contentDetails"]["videoId"]
    description = video["snippet"]["description"]
    publishedAt = video["snippet"]["publishedAt"]
    title       = video["snippet"]["title"]
    status      = video["status"]["privacyStatus"]

    #    Create a date object based on the video's timestamp
    date_object = datetime.strptime(publishedAt,"%Y-%m-%dT%H:%M:%S.000Z")

    #    Create a filepath and filename
    #    /YYYYMMDD-HHMM/YYYYMMDD-HHMM_My Video_abc123
    filepath    = date_object.strftime('%Y%m%d-%H%M')
    filename    = date_object.strftime('%Y%m%d') + "_" + title + "_" + videoId

    #    YouTube Options
    ydl_opts = {
        #    The highest quality MP4 should have the best resolution.
        #    Best quality AAC audio - because that's the codec which will fit in an MP4
        #    A lower quality WEBM for HTML5 streaming
        'format': 'bestvideo[ext=mp4]+bestaudio[acodec=aac],webm',
        'audioformat' :'aac',
        'merge_output_format' : 'mp4',
        #    Don't create multiple copies of things
        'nooverwrites' : 'true',
        #    Some videos have subtitles
        'writeautomaticsub' : 'true',
        'writethumbnail' : 'true',
        #    Make sure the filenames don't contain weird characters
        'restrictfilenames' : 'true',
        #    Add the correct extention for each filename
        'outtmpl' : filepath + '/' + filename + '.%(ext)s',
        #    If there is a problem, try again
        'retries' : '5',
        # 'verbose' : 'true',
    }

    #    Download the files
    with youtube_dl.YoutubeDL(ydl_opts) as ydl:
        if status == "private":
            #    YouTube-DL won't download private videos
            print "Skipping private video " + videoId
        else:
            print "Downloading " + videoId + " to " + filename
            ydl.download(['http://www.youtube.com/watch?v='+videoId])
            #    Write an HTML5 snippet
            html  = "<video   poster=\""+filename+".jpg\" controls >\n"
            html += "    <source src=\""+filename+".mp4\"    type=\"video/mp4; codecs=mp4a.40.2, avc1.42001E\">\n"
            html += "    <source src=\""+filename+".webm\"   type=\"video/webm; codecs=vorbis, vp8.0\">\n"
            html += "    <track  src=\""+filename+".en.vtt\" kind=\"subtitles\" srclang=\"en\" label=\"English\">\n"
            html += "</video>"
            #    Write the HTML
            with open(os.path.join(filepath, "video.html"), 'w') as html_file:
                html_file.write(html)

The files it downloads are significantly smaller than the original uploads - with no noticeable loss of quality. The combined size of the MP4 and WEBM are around half the size of the original files.

There is one major bug - occasionally the script will crap out with:

youtube_dl.utils.DownloadError: ERROR: content too short (expected 153471376 bytes and served 88079712)

This is a persistent error with YouTube-DL.

The script can be run again - it's smart enough to avoid re-downloading the videos.

And there you have it - a quick way to grab everything you've uploaded. It's missing a few things - view counts and comments, mostly - but it's good enough for re-hosting elsewhere.

This is what it looks like

This is what happens when you put that dash of HTML into a web-page:

I think there's a problem with the subtitle track - but it is not a finalised standard yet.