How to prompt Flux AI

Macro product photograph of a stainless-steel dive watch on wet black slate, water droplets beading on the sapphire crystal, dramatic single-source studio lighting from the upper left, deep shadows, subtle reflection under the case, shot on a 100mm macro lens at f/4, cold steel-blue palette with one warm highlight on the bezel.

Subject first, then material words (steel, slate, sapphire) that Flux turns into micro-detail. One named light source plus a lens spec produces catalog-grade realism without any "8K masterpiece" filler.

Fashion editorial photograph of a young woman with curly red hair and freckles, leaning against a sunlit concrete wall, wearing an oversized cream linen blazer, direct gaze, shot on Hasselblad X2D, 80mm lens, f/2.8, shallow depth of field, soft late-afternoon side light, muted warm film palette.

Camera, lens and aperture are the realism levers on Flux — "shot on Hasselblad, 80mm, f/2.8" beats "professional photo" every time. The concrete subject ("curly red hair and freckles") renders more consistently than a generic "beautiful woman".

Cinematic long shot of a lone hiker crossing a black-sand ridge in Iceland, low storm clouds tearing open above, a single shaft of sunlight hitting the valley floor below, anamorphic framing, shot on 35mm film, desaturated teal-and-slate palette with one warm accent on the hiker's red jacket.

The 21:9 ratio plus anamorphic and film-stock words gives a widescreen film-still look. One accent color against a restrained palette is a Flux strength — state both explicitly.

Studio product photo of a minimalist running sneaker floating at a slight angle over a seamless background, the sneaker upper is #0047AB, the midsole is #F5F0E6, the seamless background is a smooth gradient starting with #02eb3c finishing with #edfa3c, soft even lighting, crisp shadows underneath.

Hex codes work when each one is attached to a named object — "the upper is #0047AB", never a bare list of codes. Gradients take two hex endpoints. This is the reliable route to on-brand colors without post-correction.

Night street photograph of a narrow ramen bar entrance in the rain, the sign above the door reads "MIDNIGHT NOODLES" in warm yellow neon letters, a smaller vertical banner reads "OPEN" in red neon, wet asphalt reflecting the signs, shot on 35mm at f/1.8, steam drifting from the doorway.

Quotation marks around the exact words, plus placement ("above the door") and typography style (neon, color) — the three things Flux needs to render text correctly. Two short strings are reliable; dense paragraphs of copy are not.

A golden retriever mid-leap chasing a tennis ball across sunlit hardwood floor in a cozy living room, muddy paw prints trailing behind, a knocked-over plant in the background, warm afternoon light streaming through sheer curtains, dust motes in the light shafts, shallow depth of field, candid pet photography, 35mm lens.

An 80-plus-word prompt where every clause adds a visible element — action, evidence of action, environment, light behavior. Iterate a scene like this on Flux Dev at 1K, then rerun the winning prompt on Flux 2 Pro for the final.

Product lifestyle photo of a ceramic pour-over coffee set on a linen tablecloth, morning window light, minimal composition with generous negative space, in the brand illustration style of the loaded LoRA.

The LoRA carries the look; the prompt only supplies subject, composition and light. Load your LoRA weights by URL, keep LoRA scale at 1.0, and drop toward 0.7-0.8 if the style overpowers the subject or push to 1.2 if it barely shows.

Change the background to a sunlit Mediterranean beach while keeping the person in the exact same position, scale, and pose. Match the lighting on the subject to the new scene.

The preservation clause is the whole trick — without "exact same position, scale, and pose" the model treats the subject as editable too. Kontext edits from a single input image; upload it, then write only the change.

Replace "GRAND OPENING" with "NOW HIRING" on the storefront banner. Keep the font style, color, and banner position unchanged.

The Replace "[original]" with "[new]" structure is the documented, most effective way to edit text in an image — quote both strings exactly. Adding "keep the font style" pins the typography so only the words change.

Turn this photograph into an oil painting with visible brushstrokes and impasto texture, keeping the exact composition, subject placement, and color relationships unchanged.

Style transfer on the Flux 2 editing models works best as one clear medium instruction plus an explicit preserve list. Direct verbs ("turn this into", "change", "replace") outperform vague asks like "make it more artistic".