Flux’s prompt logic is totally different from SD’s tag-stacking style. A lot of people migrating from SD still pile on keywords like crazy, and the result just comes out muddy. Flux handles natural language way better—just write it like you’re describing a scene to someone: subject, action, environment, lighting, lens, all in a normal sentence order, and it gets it.
If you dump a bunch of comma-separated adjectives, the weights all fight each other, and it has no clue what you want to emphasize. My habit is to first write one sentence that nails the core scene, then add two or three specific constraints—like focal length, time of day and lighting, material texture.
Text rendering is where Flux beats most models; if you need signs or poster text, just put the words in quotes in the prompt. Don’t treat it like a reskinned SD—you gotta change your writing style too.