Cartoons and kids' songs: clips framed on the characters
On a YouTube video marked as made for kids, with no speech in it, Katto now frames the characters full screen. Before, it treated the lyrics written on screen like a title and kept the whole width, so the clip was a band of picture between blurred bars. Every other video is framed as before.
It started with a 2 min 26 nursery rhyme pasted on November 14. Both clips came out as a strip of cartoon between two blurred bars, 34 to 48 % of the frame tall. The detector had found the characters; what shrank the picture was the lyrics written on screen. Katto protects on-screen text like a title, so it kept the whole width, and the whole width of a landscape video is a band in a vertical frame.
The new path starts only when two things are true: YouTube marks the video as made for kids, and Katto hears no speech in it. Katto then looks for the characters twice a second (people, animals and toys at least a quarter of the frame tall), aims at the group and keeps it whole when it fits; when it does not, the main character stays whole. The framing moves smoothly and only jumps on a real cut. A shot with no character keeps the whole picture between blurred bars, and if characters are on screen less than half the time, the clip is framed as before.
What it costs: lyrics written on screen can be cropped. Re-run on November 15, the same nursery rhyme gave 2 clips of 60 and 34 seconds, both framed on the characters, and the whole job took 74.5 seconds.
2 videos a month, no card. What is still being built is on the roadmap.