If you’ve been paying mindful attention to YouTube just recently, you might have seen the increasing pattern of so-called “faceless YouTube channels” that never ever include a noticeable human talking in the video frame. While a few of these channels are merely authored by camera-shy human beings, a lot more are completely automated through AI-powered tools to craft whatever from the scripts and voiceovers to the images and music. Unsurprisingly, this is frequently offered as a method to make a fast dollar off the YouTube algorithm with very little human effort.

It’s not difficult to discover YouTubers grumbling about a flood of these faceless channels taking their ingrained records files and running them through AI summarizers to produce their own immediate knock-offs. One YouTuber is attempting to combat back, seeding her records with scrap information that is unnoticeable to human beings however toxic to any AI that attempts to attempt to work from a poached records file.

The power of the.ass

YouTuber F4mi, who produces some exceptional deep dives on unknown innovation, just recently detailed her efforts “to poison any AI summarizers that were trying to steal my content to make slop.” The essential to F4mi’s approach is the.ass subtitle format, developed years earlier as part of fansubbing software application Advanced SubStation Alpha. Unlike easier and more popular subtitle formats,. ass supports expensive functions like typefaces, colors, placing, vibrant, italic, highlight, and more.

It’s these expensive functions that let F4mi conceal AI-confounding trash in her YouTube records without affecting the subtitle experience for her human audiences. For each portion of real text in her subtitle file, she likewise placed “two chunks of text out of bounds using the positioning feature of the .ass format, with their size and transparency set to zero so they are completely invisible.”

In those “invisible” subtitle boxes, F4mi included text from public domain works (with particular words changed with synonyms to prevent detection) or her own LLM-generated scripts filled with totally fabricated truths. When those records files were fed into popular AI summarizer websites, that scrap text wound up frustrating the real material, producing an absolutely unassociated script that would be worthless to any faceless channel attempting to exploit it.