Maybe, but in my limited experience keyword prompts work better today. You need to describe many orthogonal aspects of the image at once within a small token budget, so brevity counts. Also, numerically weighted keywords are more precise than saying "very" etc. in prose.

→ View original post on X — @goodside