Why image models can't spell, and how to overlay text with Canvas instead

Image models produce artwork well but frequently misspell words, which is a serious problem for covers, posters and jackets. One design team therefore has the model paint text-free artwork and draws the artist name and title in the browser using the Canvas 2D API. This makes the lettering reliably correct, allows edits without regenerating the image, and lets them use a cheaper image model. Implementation required loading web fonts explicitly through the FontFace API so Canvas does not silently fall back to a default face. They also fit titles by trying sizes from largest to smallest and checking every wrapped line, which fixed a bug where an overlong word overflowed a flyer.