The short version
Generic AI images scream 'generated'. The fix is making the media agent read the article first — tone, structure, style — before it draws.
The art-direction bottleneck
Every content site hits the same wall: writing scales faster than art direction. The result is either articles with no imagery, or a feed of interchangeable gradient-hero images that readers have learned to ignore.
Better inputs before better prompts
The fix isn't better prompts, it's better inputs. Before generating anything, a media agent should read the article it's illustrating: the argument, the section structure, the tone (technical? editorial? playful?), and the site's existing visual conventions. The image spec falls out of the reading — a how-to gets step diagrams, a comparison gets a side-by-side visual, an opinion piece gets one strong hero and no filler.
Accessibility is part of the asset
Alt text deserves the same care, because it's doing double duty: accessibility for readers and context for search engines. Good generated alt text describes the specific image, works the primary keyword in naturally at most once, stays under about 125 characters, and is never duplicated across images in the same article.
A reviewable media workflow
In SEO Forge, the writer embeds an image specification — prompt plus alt text — at every point in a draft where a visual earns its place. The image generator honors those specs exactly, flags anything it had to skip, and returns a labeled manifest so a human can review every asset before it ships.
Scale without sameness
Scale without sameness is the goal: a hundred articles, each illustrated like someone actually read it. Because something did.