Promptelle turns reference images into AI prompt drafts

The article frames prompt writing as an observation exercise rather than a vocabulary exercise. Its first recommendation is to split a reference image into what is present and where it sits, giving the example of a portrait where you specify the number of people, the impression of their age, their gaze, whether the shot is upper body or full body, and the camera distance. Specifying only an art style while leaving composition vague, it says, makes the generated layout hard to stabilize.

The second and third recommendations build on that base. Light and background should be described from observable facts—the direction of the light source, its hardness, the time of day, and the depth of the background—so that abstract phrases like "beautiful lighting" become concrete observations such as soft side light coming through a window or a background blurred with shallow depth of field. Only after subject, composition, light, and background are organized should mood be added, through things like a warm-toned film look, natural skin texture, or restrained grain.

The article then presents Promptelle as a starting point for people who find it hard to put an image into words, noting that its image-to-prompt generator can produce a draft from a reference image that users then edit. Its advice for refinement is incremental: after reviewing the output, add only the missing elements one at a time. That habit, the article suggests, makes it easier to see which instruction affected the result and to grow the prompt into something reusable.