A few starting points
Each one fills the form. It does not lock anything. Clear a slot afterward if you do not want it.
Series
Choosing a series fills the year, the decade, the scan size, and the character designer. The series name itself is written after the character names. Studio is shown for context and is never added to the caption.
Prefix
Every training caption started with anime screenshot. That token means a frame from a film or a television print, not a modern illustration. Leave it on unless you are testing what happens without it.
Year and decade
These sit immediately after the prefix. The decade is the ten-year label, so 1985 also writes 1980s. The training window is 1977–1989. Omit the year and the decade goes with it.
Characters
Names are written before the series, in lowercase. Training only named a person when that face was identified, so an empty cast is a normal still. Count tags such as 1girl are description tags, not names. They are suggested, never added for you.
Description
The open middle of the caption. Training used either a tag list or one sentence for this slot, not both. Tags are written in a fixed reading order, not the order you click them.
Scan size
hires means the source was 720p or taller. lowres means it was not. The token is written after the description and before the designer. A 480p television frame should stay lowres so the model does not treat tape softness as the era itself.
Character designer
Written after the scan size. The automatic choice is the animation character designer, not the original novelist or manga artist, when the catalog made that distinction. Vampire Hunter D writes noriyasu yamauchi, not Yoshitaka Amano. Pick someone else, or leave the slot out.
Quality
These come last. They separate a clean cel from tape noise, so the decade does not have to mean VHS. Hires stills were tagged hq cel animation by default. Broadcast faults such as vhs and dot crawl were kept for lowres captions and stripped from hires ones.