it's a music anime douga world

こなた達でアッーウッウッイネイネ【らき☆すたMIX】 by lucky #sm3913273
[image or embed]

— MAD World Service (Bot) (@mad.denden.garden) Feb 22, 2025 at 1:35 AM

no, it doesn’t actually stand for that. isn’t the lie so much better, though?

it is difficult to think people would ever want to read about MADs from the perspective of plunderphonics, though I still intend to attack it from that angle at some point. it is much easier, however, to imagine they may want to see gibara yelling like a monkey attacking dongs.exe set to ultra super hyper miracle romantic. it’s a mad world out there, and we all choose to vent our frustrations in different ways. now, you can immerse yourself in the cultural chaos of the moment by following the MAD world service, currently posting 30 second clips sourced from nico.

with vocacolle currently in full swing, this idea was born out of wanting to promote song entries beyond link dumping. vocacolle as an event, though, is primarily oriented around the three day rankings, so it doesn’t necessarily provide much of a window to circulate links. MADs are a natural alternate, and we’re not starved for choice or longevity with them. vocacolle might instead benefit from a strict ranking bot, similar to the border ranking bots done for many soshage, and vocaloid overall could make more sense with a general approach pulling down PVs from vocaDB.

nico has lost its appetite for open culture, which has seen it retreat back inwards after it once had broader global ambitions. MADs are, conceptually, one of the purest expressions of this, and they fly in the face of copyright norms while also not depending on being a market substitute to succeed. for how pervasive they are considered now, MADs primarily home to nico already once belonged to an era where they required mirroring to external services to be seen globally, until a login no longer became required to view videos in 2018. efforts like the idolm@ster MAD world service, started in 2008 by japanese producers, were born to bring some of the charm for the MAD to english audiences by explaining trends and relationships in them, when this restricted access was the status quo. with regional balkanization again becoming a reality for many services, it is incumbent to liberate access where we can. circulating kichigai tapes is the heritage of the format, and this bot is intended to be as much a part of the same lineage.

atproto has great integration for bots now that the API has matured and we’ve started seeing libraries pop up providing the glue to interact with it. jetstream also has made it much easier to interact with the firehose if you need to collect cute girls¹ or start arguments. while mastodon has been a sanctuary for good bots for as long as twitter has threatened revoking API access, the shutdown of the botsin.space instance especially has created a vacuum. with bluesky’s greater growth, and other large services like reddit also now charging bots for the privilege, atproto at the moment seems like the only approach worth promoting with JAV titles and DMV descriptions.

the glue here is fairly simple, contained to one python script. using mostly my personal pinboard bookmarks, video IDs were ingested into a sqlite database. every three hours, a cron task is kicked off to select a random row in the lone videos table, fetch the video and its associated metadata with yt-dlp, and then splice a randomly positioned clip from it using ffmpeg. that video then gets uploaded to the account’s repo as a blob, which can eventually be served as video. this is a sad path alternative to using the newer uploadVideo endpoint, which allows you to track the conversion job before actually posting, but it’s what’s already supported in the python atproto SDK. as a result, new posts may initially show an invalid video embed for a few seconds until a server gets kicked to perform the conversion. future iterations could involve something like puppeteer to also capture the comments danmaku, but having screenshared niconama before, I can already expect that people would only ask for them to be turned off. tasteless, if you ask me.

right now I’ve set the rhythm at one post every three hours. currently there are about 1500 entries ingested in the database, which should give us about half a year of runway before we need to dig deeper. these weren’t verified, but past analyses suggest we should expect somewhere around 10-15% of them to be deleted. some sort of heuristic pulling from the site’s rankings will probably come into play after exhausting this set. nico reports over 150,000 active videos tagged with otoMAD as of this writing, so I’m not anticipating it to be a problem when we need to expand. we can also get into the weeds of pulling from my own download archives, if we ever desire to tread that deep. konami’s going to be thrilled with all the yu-gi-oh MADs I’ve been holding on to after they were last pillaged with takedowns.