How to write an AI image prompt
Published
Start with a structure
A prompt is just a description — but the makers of FLUX, Black Forest Labs, recommend giving one a shape: subject, action, style, context. The subject is what the image is about; the action is what it is doing; the style is the look; the context is everything around it.
"A penguin wearing a tiny tuxedo" is a subject plus a detail. Add an action ("attending a wedding") and a style ("documentary photo"), and the model has a full sentence to work from. There is no single correct format — what matters is giving the model enough to work with.
A penguin wearing a tiny tuxedo to a penguin wedding
Try this prompt →Borrow the language of photography
The official guide's tip for believable realism: describe the image as if a real photograph were being taken in the moment. Name the lens, the lighting, the framing — and explicitly ask for real texture like pores, fabric wear, or imperfections.
Compare "a swan on water" with the version below: still water, soft natural light, a visible reflection, out-of-focus foliage as a natural vignette, warm late-day tones. Every clause steers the render toward something that reads as a photo.
Realistic moody nature photograph of a white swan gliding calmly on still dark water, captured in soft natural light. The swan's reflection is clearly visible on the surface. The scene is framed by out-of-focus foliage in the foreground, creating a natural vignette and a sense of depth. Warm golden tones from late-day sunlight.
Try this prompt →Refine one detail at a time
Strong prompts usually come from iteration, not from writing the perfect prompt on the first attempt. The guide's loop: start simple, check what the model got right and wrong, then adjust one important detail at a time — subject, framing, lighting, or style — before rewriting everything.
The guide demonstrates this with a man in a charcoal wool coat that gains a wet night street, then editorial lighting, then a teddy bear, then parked vintage cars. Each pass changes one thing, so each pass teaches you something.
You don't have to write in English
FLUX models can be prompted in many languages and respond with the same level of quality — the guide calls this out explicitly. A French market scene can be described in French, and the model handles it natively.
For Chinese prompts specifically, Z-Image-Turbo is tuned for bilingual scenes — see its model page for official Chinese example prompts.
Un marché alimentaire dans la campagne normande, des marchands vendent divers légumes, fruits. Lever de soleil, temps un peu brumeux
Try this prompt →See it applied
Every item in the curated gallery is an official example pair — the published image next to the exact prompt that produced it — each with a one-click hand-off into the generator so you can remix it with any model on the site.