Images were extracted from the 100% TotK Playthrough from 31json. As the videos were 10s of GB in size, downloading them became prohibitive.
Times when Koroks appeared in the video along with their IDs was completed, slowly, using a basic video notes tool. The video included the route displayed in notceler-totk. Without the route, it would have taken much longer to identify the Koroks.
Once we have the times and the youtube video ID we then use yt-dlp to extract a desired 15 seconds of video surrounding the Korok. This uses the --download-sections *t1-t2 option. The clips are normally 20-25 seconds due to keyframes in the video encoding.
Using the videos, we then extract an image at an estimated time using ffmpeg with the options for cropping, -vf crop:$W:$H:$X:$Y and grabbing a single frame -frames:v 1.
At this point, we use the video to identify the "best" image of the korok, just before they say "Ya-ha-ha!", and the leaves have blown away. Remake the images from the videos and we are done. This was done using specialized html hiding in the page that is quite similar to the note taking tool.
Images of the Korok Friends were kindly created by 31json on a specific playthrough. Each Korok Friend was visited and reunited. Hurrah! Boxes, wooden or metal, from Amiibo were spawned and placed behind Link to make him disappear using specific camera positioning. The HUD was also turned off to provide a cleaner image.
This is substantially different than the BotW Koroks. Which was completed mostly automatically using cross-correlation in the audio stream. Ya-ha-ha! is rather distinctive.
2025-01-06 - savage13 - Written because to remind myself how this was done.