I’ve been describing some of the new site design details in a variety of threads here. I started with super-technical posts since that’s where my focus has been (you need to get the concrete base poured before you can build the walls of the house). But recently, I’ve started describing how site members will use some of the technical features. Unfortunately, this has caused some unease because the overall process and user interface is not visible, leading to concerns that the new process and interface may be overly complicated.
To help put minds at ease, I’ll describe the vision for the new process.
Where are we now?
The site as you see it now is built on technology, interfaces, and methods that were cutting-edge 20-25 years ago. Obviously, things have changed significantly since then, and the site has not kept up with the times. Compared to modern methods, the current site blows big-time. I created the site, the site is my baby, but I’ll be the first in line to tell you that my baby is ugly... like... eeewww...
What’s the plan to make things better?
A popular modern video platform is YouTube. The process to upload a video there is basically: sign in to YouTube, click “Create” -> “Upload video.”, upload the video file, enter a title, set the “Kids / Not Kids” option, set visibility to “Public”, and click “Publish”.
The Org’s goals, as far as AMVs go, have a few more aspects and thus a few more steps involved. But, let’s see how the upcoming new method compares to the old method and a modern interface while keeping in mind the site’s core mission: catalog, archive, and educate.
Using the new Org AMV entry method, the ideal duration for an existing site member to make a newly uploaded AMV available to the public while still providing all proposed minimum required information (entries for anime, music, standard tags, etc) is about 5 seconds after upload. Yup, that's right: 5 seconds. Let’s walk through the ideal process.
Let's say you are Kaylee Thompson and you want to create an entry for your brand new “The Pony Man” AMV (one of my favorite AMVs ever, btw). After logging into the site, you click the “New AMV” button. You are presented with a dialog box that says “Drag the AMV file here to upload or click [here] to add a new entry without a video file”. You drag and drop your “The Pony Man.mp4” video file and wait for the upload to complete.
Once the upload is done, the site shows a message, “Upload complete! Autodetecting video information…” and a twirling graphic appears with words underneath it that constantly change: “Detecting file attributes…”, “Detecting music…”, “Detecting anime…”, “Compiling results…”
You are then presented with a newly filled-in video entry that looks like this:
- Creator: Lostgirl (Kaylee Thompson)
- Title: The Pony Man
- Premiere Date: (today’s date and time)
- Song: “The Pony Man” by Gordon Lightfoot
- Anime: My Neighbor Totoro
- Description: (blank)
- Content Descriptors:
- Supernatural = Low, several, clearly depicted;
- Tone = Light, frequent, clearly depicted;
- Specific Warnings: None
- Tags: None
- Content Rating: Y / Kids
- Duration: 3:41
- Availability: Streaming and download
The AMV entry is now available for general consumption. Fewer steps than YouTube!!
To recap, the ideal “add new AMV” process looks like this:
- Log in
- Click the “New AMV” button
- Drag-and-drop the AMV file
- Wait for the upload and autodetection to complete
- View autodetected information (ideally everything is accurate)
- Click “Publish”
That’s impossible! What magic is this??
Yes, it seems like magic, but there is proven technology we can use to get all of this done.
Starting with the easy parts of the process: The creator is you since you’re logged in. The title is autodetected from the filename (no extension). The premiere date is just today. The duration is detected as a property of the video file (a command-line tool to detect it). But next comes the more interesting parts.
To detect the anime, we will push each frame of the video through a local copy of trace.moe, a scene-based anime detection tool. The tool works by comparing the video frame with a known library of video frames. The creator of trace.moe has scanned thousands of anime and publishes the detector results along with the detection tool itself for free for anyone to use. The output of each scanned frame is the anime, season (if a series), episode, and approximate timestamp at which the frame occurs in the source material.
There are similar online tools for detecting music. Unlike scanning an entire video frame-by-frame to determine all possible anime (a 3-minute AMV at 30 fps has 5,400 frames to scan), identifying a song is a one-step process since an AMV typically has only one song in it. Stripping out just the audio into an MP3 and uploading that small file to a music identification service is easy to automate, masks the source of the audio, and returns a good amount of metadata about the song.
Now that the anime is listed and the song is identified, we can match those detected to our own standardized databases of anime and music to add the centrally located common content descriptors.
But that’s ideal. What is the typical upload experience going to look like?
The typical process will look much like the above, especially with the easy-to-detect aspects. However, you may need to adjust the title and premiere date manually, especially if you want to premiere the video in the future but upload it now. People usually like to see a description, so adding that is a manual step. You will be allowed to add a video cover photo for the video entry, and a good number of creators do this.
With more than 80% of AMVs having only one or two anime in them, it is extremely likely that the correct anime will be auto-selected as a source. However, there may be a false positive in the list (an anime that was autodetected incorrectly). In that case, the uploader would just need to uncheck the anime that isn’t actually in the AMV before publishing.
But if you’re an advanced editor who visually modifies every frame of your AMV, the current autodetection method has a much lower chance of returning results. In that case, you’ll have to add the anime manually. But even then, it only takes a single frame to match an anime to return a result, so this challenge will be very few and far between.
While anime and music sources that are benign will easily put “none” for the standardized content descriptors, default suggested descriptors may be too severe for your particular AMV compared to the overall series content warnings (I’m looking at you, “Made in Abyss”...) and may need to be adjusted down to fit your particular edit focusing on non-objectionable material within an otherwise strongly rated source. The good part is that adjusting down takes much less effort than adding the indicator in the first place.
If your AMV has multiple songs in it, the song detector may or may not successfully detect the music. In that case, you’ll have to enter the music manually.
Tags are another detail that creators like to add. Currently, auto-assigning tags will be blunt and may not fit your edit. But since the anime autodetection returns the detected anime down to nearly the exact frame used from the source, tagging specific scenes from specific anime could be added at a later date to make auto-tagging much more accurate. But even then, adding your own specific tags should be an efficient but manual effort.
The autodetected content rating (G, PG, PG13, etc., still under development) will be the cross product of the anime and music base ratings (meaning a benign anime with an explicit lyrics song will auto-assign a harsher rating), along with the modifiers associated with tags. However, this cross product may be too lenient for highly creative editors. But, adjusting this indicator should require little effort.
Categories are the last part that would require manual entry. I currently can’t think of a way to auto-suggest an AMV category. But if someone can think of a way to detect it, we can implement it to fill in that blank. That last thing is availability: if you want to adjust that, it will take about two clicks or so.
Overall, the autodetection system should reliably prefill the majority of metadata content (other than description and video entry image) and allow easy adjustment of inaccurate values.
But you know what? Given the right circumstances, the whole process can be even easier.
Wait, it can be even easier?
Yes. As mentioned above, a modern video hosting platform is YouTube. Let’s say you’re a long-time editor who entered your videos for years from 2009 until about 2017, when you finally had enough of the 2002-era upload process and just uploaded your videos to YouTube and that’s it. Now it’s been years since adding an entry at the Org and you’re about 37 AMVs behind in documenting your catalog. The thought of manually entering that many AMVs in one sitting is just.... uuuuuugh......
Lucky for you, there are ways of automatically migrating video entries and video files from YouTube to here. You could provide the site with your link to your YouTube entry and the site would get the basic details (title, date premiered, description, thumbnail, tags, etc) and the video file in its highest quality form. The site would then use the above automated processes to fill in the blanks of anime, music, content descriptors, and so on.
But even better, you could provide the link to your YouTube channel as a whole. The site could then read off all video entries in that channel, autodetect which ones have anime in the video, and create draft entries for those (including the video file). All of this could happen in the background without further direct interactions, regardless of the number of entries in the channel. You would just need to confirm the information before publishing the entries on the Org.
As a safety measure against impersonators, the Org would require you to authenticate with Google before digesting an entire YouTube channel. Once authenticated, the channel would only be associated with your own Org account once, meaning if any other Org account attempted to associate with your YouTube account, they would be autorejected.
This means that you could document your entire catalog here with only a few clicks and safeguard against take-downs from YouTube (which happens fairly often).
So, that’s the concept of the revamped AMV entry system in a nutshell (if the nutshell was as big as a Volkswagen).
Does this help clarify the vision of the new AMV upload system? Do you feel more at ease since a large amount of previously manual effort could be automated?
Again, I greatly encourage feedback, both concuring and concerning.
Thanks again and have a great day!!
Phade.
)