These past two days, I’ve been making X image cards, and I changed one thing: don’t let the image model handle Chinese characters.

Image models are good for composition, texture, and lighting; the spacing, line spacing, and information hierarchy of Chinese characters still need to be laid out manually.
Now I only use Image2 as a style reference, and I do a final check with a 320px feed preview.
It’s not enough for the card to look good in the original image—it also needs to be understandable at a glance as you scroll past it.