You type “beautiful cinematic lighting, masterpiece, ultra detailed” and get a soft, half-melted face. Same person, rewritten as subject first + lens + light source, and the skin, eyes, and wardrobe suddenly hold. That gap is usually structure, not luck.
This note covers how to write FLUX prompts the way Black Forest Labs’ FLUX.2 prompting guide describes them—plus one reusable formula and 40 copy-paste FLUX prompt examples you can drop into a text-to-image run as-is.
Rules that actually move FLUX results
These track the official FLUX.2 guidance (subject-first order, no negative prompts, medium-length prompts, camera/hex specificity). Community habits are labeled as such.
1. Lead with the subject
FLUX.2 weights early tokens more heavily. Put the person, product, or place first.
- Weak: Soft dreamy portrait with beautiful lighting, elegant woman
- Stronger: Korean woman, 28, short black bob, looking past camera, 85mm studio portrait, soft key from camera-left
2. Name camera, lens, and light—not “cinematic”
Official photorealism tips prefer concrete gear and light over vague praise words.
- Weak: Professional photo, cinematic lighting
- Stronger: Shot on Canon EOS R5, 50mm f/1.8, window light from the left, shallow depth of field
3. Aim for medium length (roughly 30–80 words)
BFL’s guide calls medium prompts “usually ideal.” Short is fine for exploration; long helps only when the scene is genuinely complex.
- Weak: coffee mug product shot (too thin) / a 200-word mood essay (often muddy)
- Stronger: Matte black ceramic mug on polished concrete, thin steam, three-point softbox, 85mm, f/5.6, clean commercial product photo
4. Do not rely on negative prompts (FLUX.2)
Official note: FLUX.2 does not support negative prompts. Say what should be there instead of “no blur / no people.”
- Weak: no blur, no people, no watermark
- Stronger: sharp focus on the bottle label, empty white cyclorama, clean frame edge to edge
5. Bind hex colors to objects
Hex works best when tied to a named part (“the hoodie is #FF5733”), not floating in the prompt.
- Weak: use #0A84FF somewhere
- Stronger: aluminum water bottle body in color #0A84FF, brushed steel cap, white seamless backdrop
6. Put style anchors after the subject
Style still matters; it just should not bury the subject.
- Weak: anime, best quality, 8k, girl in rain
- Stronger: Teen girl under a clear umbrella in Tokyo rain, cel-shaded anime illustration, bold ink outlines, muted teal palette
7. Quote on-image text and place it
FLUX.2 can render readable type when you quote the string and describe placement/style. Success is still inconsistent—plan to regenerate.
- Weak: poster that says something about coffee
- Stronger: vertical poster; large headline text "BREW SLOW" in bold condensed sans, centered top third, cream paper texture
Community experience (not an official claim): stacking Midjourney-era quality tags (masterpiece, 8k, trending on ArtStation) often adds noise on FLUX-family models; swapping them for materials + lens + light usually stabilizes faces and products faster.
A reusable FLUX prompt formula
Main formula (natural language)
[SUBJECT + key traits] + [ACTION / POSE] + [SETTING] + [CAMERA / LENS / LIGHT] + [MATERIALS / COLORS] + [STYLE ANCHOR]
Sensitive for FLUX / FLUX.2: subject identity, quoted text, hex tied to objects, camera+lens, light direction.
Often optional: mood adjectives, “highly detailed,” quality spam, artist name piles (unless you truly want that look).
Variant A — photoreal product / portrait
[exact subject], [pose], [surface/background], shot on [camera], [lens], [f-stop], [light setup], [material + hex if needed], commercial photo
Variant B — complex / brand-controlled (JSON, per BFL)
Use when multiple parts need independent control. FLUX.2 accepts JSON or flattened prose; pick based on workflow (JSON prompting guide).
{
"scene": "studio product setup on matte acrylic",
"subjects": [
{
"description": "glass serum bottle with frosted label",
"position": "center",
"action": "standing upright"
}
],
"style": "commercial beauty photography",
"lighting": "soft overhead softbox, gentle rim from behind",
"camera": { "angle": "eye level", "lens": "90mm", "depth_of_field": "label sharp, background soft" },
"color_palette": ["#F5F0E8", "#1A1A1A", "#C9A227"]
}
Open-weight context if you local-run older FLUX.1 checkpoints: see the FLUX.1 [dev] model card on Hugging Face. Licensing and commercial terms differ by checkpoint—check the official page for the build you use.
40 copy-paste FLUX prompt examples
Paste these as written, then change one variable at a time. For multi-model checks, an AI image generator overview helps compare behavior; prompts below are tuned toward FLUX / FLUX.2 habits.
Product / ecommerce
Matte ceramic pour-over dripper in color #2F2F2F on light oak, single drip of coffee mid-air, soft window light from the left, shot on Sony A7IV 90mm f/2.8, seamless pale gray backdrop, ecommerce catalog photo
Catalog hero; fails when you omit material + hex and only say “nice coffee gear.”
White wireless earbuds in an open charging case, case lid color #E8E8E8, buds color #111111, top-down flat lay on white marble, even softbox lighting, 50mm, sharp edges, minimal product photo
Top-down SKU; watch for melted mesh grills if you skip “sharp edges.”
Trail running shoe, neon midsole color #CCFF00, black knit upper, three-quarter view on seamless white, hard key light with soft fill, commercial footwear shot, dust-free
Footwear; “dust-free” beats “no dust” on FLUX.2.
Glass olive oil bottle with green glass tint, paper label reading "OLIVA", cork stopper, rustic stone counter, late afternoon side light, 85mm food-product photo
Label text may need a few seeds; keep the quote short.
Skincare dropper bottle, frosted glass body color #F7E7D3, gold metal collar color #C9A227, standing in shallow water with soft ripples, beauty advertising still, soft overhead diffusion
Beauty ads; overdoing “bokeh bokeh” can blur the bottle.
Foldable laptop stand, aluminum body color #A8A8A8, partially folded, isometric angle on charcoal desk mat, clean desk product render look, crisp reflections
Tech accessories; “isometric” helps geometry stay sane.
Stack of three paperback novels bound with twine, top book cover color #B22222, soft morning light, shallow depth of field, lifestyle product photo on linen cloth
Lifestyle stack; name which book is on top.
Ceramic plant pot color #D9C4A5 with a small monstera, centered on white cube pedestal, soft studio light, plant ecommerce listing, leaves sharp and healthy
Plant SKUs; “healthy leaves” reduces yellow wilt artifacts.
Portrait / character
South Asian man, 35, salt-and-pepper beard, navy linen shirt, seated half-turn toward camera, shot on Hasselblad X2D 80mm f/2.8, soft north-window light, neutral gray seamless, editorial portrait
Headshots; vague “handsome” often softens facial structure.
Black woman, 24, natural afro, gold hoop earrings, laughing with eyes half-closed, outdoor café patio, golden hour backlight with soft fill, 35mm street portrait
Candid joy; “eyes half-closed” avoids frozen grin faces.
Elderly Japanese potter, clay-stained apron, holding a half-finished bowl, workshop shelves soft in background, Kodak Portra 400 look, warm practical lamps, documentary portrait
Character work; film stock anchors tone better than “vintage vibe.”
Non-binary model with buzzed hair and freckles, oversized denim jacket, looking down at hands, studio rim light from behind, soft front fill, fashion lookbook frame
Fashion; specify gaze or hands idle oddly.
Fantasy ranger, mid-20s, wind-burned skin, leather hood, short bow across back, standing on a misty cliff edge, overcast light, photoreal character concept, 50mm
Game concept; keep costume materials concrete.
Child, about 7, red raincoat, holding a paper boat, puddle reflection, rainy suburban street, early digital camera candid look, soft flash fill
Kids; keep pose simple—complex hands fail more.
Cyberpunk courier with scratched helmet visor half-up, neon alley reflections on jacket, night rain, teal and magenta practical lights, cinematic still, sharp facial detail under visor edge
Sci-fi; “sharp facial detail under visor edge” fights mushy faces.
Ballet dancer mid-leap in black leotard, empty rehearsal studio, side spotlights cutting dust, frozen motion, 70mm, high shutter feel, performance portrait
Motion; name “mid-leap” once, don’t stack verbs.
Environment / landscape
Coastal lighthouse on a rocky headland at blue hour, wet basalt, long-exposure feel on the sea, cool fog, shot on Fujifilm GFX, wide 24mm landscape photo
Landscapes; “blue hour” beats “epic sky.”
Quiet cedar forest path after rain, moss on roots, soft volumetric shafts through canopy, desaturated greens, 35mm nature photo, no people
Empty scene: say “no people” as empty path, or “uninhabited trail”—prefer “uninhabited trail, empty path.”
(Rewritten for FLUX.2-positive style:)
Quiet cedar forest path after rain, moss on roots, soft volumetric shafts through canopy, desaturated greens, 35mm nature photo, uninhabited trail, empty path
Narrow alley in Lisbon, azulejo tiles, hanging laundry, midday hard sun and deep shadow, documentary street photo, 28mm
Travel; name the city cue once.
Frozen lake under clear night sky, faint aurora on the horizon, lone wooden dock in foreground, long exposure star points, cold color grade, 20mm
Nightscapes; keep aurora “faint” to avoid neon soup.
Desert mesa at midday, wind-carved sandstone, sparse scrub, high-contrast sun, National Geographic field photo feel, deep depth of field
Arid scenes; “deep DOF” keeps rock texture.
Tokyo residential street at 6am, vending machines glowing, wet asphalt, quiet atmosphere, 2000s digicam style with slight noise, candid urban frame
City mood; digicam cue is in BFL’s style table.
Alpine meadow in late spring, wildflowers in foreground, snow peaks distant, soft overcast light, 50mm landscape, calm composition
Foreground flowers need “in focus” if they matter.
Abandoned greenhouse with broken glass panes, overgrown ferns inside, shafts of dusty light, muted teal and rust palette, architectural ruin photo
Interiors; “dusty light” > “creepy vibes.”
Poster / graphic with readable text
Readable text is better than older models on FLUX.2, but still not guaranteed. Keep strings short; regenerate seeds.
Vertical concert poster, large headline text "NIGHT BUS" in heavy condensed sans, small subtext "LIVE AT THE DOCK", dark blue field color #0B1C2C, orange accent bars color #FF6A00, clean vector-poster layout
Music posters; long lyrics lines break more often.
Square social ad, bold text "FREE TRIAL" centered, supporting line "7 days", product bottle on the right third, white background, modern DTC ad layout, high contrast type
Ads; two short strings beat a paragraph.
Magazine cover mockup, masthead text "FIELD NOTES", cover line "Maps for Slow Travel", portrait photo of a hiker below, classic serif masthead, print-ready editorial cover
Covers; quote masthead exactly.
Cafe window decal design on glass, arched text "OPEN DAILY", small text "8AM–6PM", frosted vinyl look, street reflection soft behind glass, signage photo
Signage; curves increase typo risk.
Minimal book cover, title text "SILT", author text "M. CHEN", textured paper color #EDE6DA, single pressed leaf motif, quiet literary design
Books; one-word titles land more often.
Infographic-style square, title text "Water Use", three icon rows with short labels "Home", "Farm", "Industry", palette #0E4D92 and #E8F1F8, clean corporate infographic
Infographics: short labels only; dense stats often garble.
Neon storefront sign reading "NOODLE BAR" in red neon letters above a steamed-up window, rainy night street, photograph of signage, readable neon
Neon; quote the words; expect occasional letter swaps.
Packaging front panel mockup, brand text "SOLAR SALT", subtitle "Coarse Flakes", kraft box color #C4A484, centered typography, studio packshot
Packaging; keep brand + subtitle only.
Stylized / illustration
Isometric tiny bookstore interior, warm lamps, cats on ladders, soft cel shading, limited palette of cream and brick red, cozy game-art illustration
Isometric; “limited palette” reduces muddy gradients.
Linocut print of a fishing boat, thick carved lines, two-color ink on cream paper, handmade print texture, poster-sized composition
Print styles; name the medium early after subject.
Watercolor botanical study of a passionflower, pencil underdrawing faintly visible, uneven paper wash, natural history plate style
Botanicals; “study” keeps it flat and readable.
Flat vector city map of a fictional harbor town, labeled pier and market, pastel blocks, clean iconography, board-game map illustration
Maps; too many labels → gibberish; keep 2–3 names.
Ink-and-wash samurai standing in bamboo rain, expressive brush strokes, high-contrast black ink, traditional sumi influence, vertical scroll composition
Ink work; avoid stacking Western + Eastern style tags.
Retro travel postcard illustration of Reykjavik harbor, bold mid-century shapes, limited six-color print look, caption space at bottom left left blank
Postcards; “caption space left blank” if you don’t want fake type.
Cute chibi robot watering houseplants, chunky shapes, soft plastic materials, pastel desk scene, children’s sticker-sheet style
Cute styles; “chunky shapes” fights over-detail mush.
Dark fantasy tarot card frame, central figure a lantern-bearing fox, ornate border, muted gold and indigo, engraved illustration style, no readable microtext
Ornate frames; ban microtext by describing a clean border.
Debug checklist (one change at a time)
- Freeze the seed (if your UI allows) once a composition is close.
- Change only one of: subject → camera → light → style → framing.
- If faces smear: shorten the prompt; move identity traits to the front; add lens + light.
- If brand color drifts: attach hex to the named part; drop unrelated palette words.
- If text fails: shorten the string, quote it, state placement; try another seed before rewriting the whole scene.
- If the scene is overcrowded: remove secondary props; or switch to JSON so each subject has a position.
- When to change model, not the prompt: you need stricter typography, a different anatomy bias, or a style the current checkpoint keeps resisting after 5–8 structured tries. Prompting can’t fix a model mismatch. License/commercial terms: follow the official page for that model.
If you’re restyling from a reference still, image-to-image is the cleaner path than stuffing “in the style of my photo” into a pure text prompt.
FAQ
Do FLUX prompts need negative prompts?
On FLUX.2, official docs say negative prompts are not supported. Describe the desired state (“sharp focus,” “empty street”) instead of bans. Other UIs may still show a negative box for older stacks—behavior depends on the exact pipeline; trust the docs for your model.
How long should a FLUX prompt be?
BFL’s FLUX.2 guide suggests short (10–30 words) for exploration, medium (30–80) as the usual sweet spot, and long (80+) when the scene truly needs it.
Why does “cinematic lighting” keep failing?
It’s underspecified. FLUX responds better to light direction, quality (softbox vs hard sun), and camera/lens. “Cinematic” alone doesn’t pin a look.
Can FLUX.2 render logo text reliably?
It can produce readable type when you quote the string and specify style/placement, but results vary. Keep copy short and plan multiple seeds. Don’t promise pixel-perfect brand lockups from one pass.
Natural language or JSON?
Both are supported on FLUX.2. Natural language for speed; JSON when you need independent control over parts, colors, or automation.
Closing
Write FLUX prompts as instructions: subject first, then action, then concrete camera/light/materials, then style. Skip negative lists on FLUX.2; quote short text; bind hex to objects. Iterate one variable at a time—and switch models when the failure mode stops being linguistic.