Free FLUX.2 Prompt Generator (Photorealism Presets)

TL;DR: A FLUX prompt generator is a fill-in-the-blanks brief plus two pixel dimensions. FLUX.2 takes width and height, not an aspect-ratio string, and has no negative-prompt field on any endpoint. Fill the seven slots below, pick a variant, and copy one of 25 photorealism presets.

Everything numeric on this page was checked against Black Forest Labs' own documentation and OpenAPI spec on August 27, 2026, and each section names where the figure came from. If you want the optics behind the lens numbers rather than the fill-in version, our guide to camera and lens terms that change your AI images covers those. This page is the tool.

What does this FLUX prompt generator actually produce?

Seven slots and two numbers. That is the whole thing.

FLUX.2 parses sentences, not tag weights. BFL's prompting guide describes the model reading a clear description of the image you want, and its own worked example builds a prompt by adding image type, then subject, then location, then visual direction, then colour and grain, one layer at a time. The generator below is that sequence turned into a form you can fill in ninety seconds.

FLUX.2 PHOTOREALISM GENERATOR — fill every slot, then delete the labels

CAPTURE:   [what kind of photograph this is: editorial portrait, e-commerce
            packshot, street documentary frame, architectural interior]
SUBJECT:   [what it is, plus two physical details a stranger could not guess]
SETTING:   [what is behind it, what is under it, how far away that is]
LIGHT:     [source + direction + quality, and what it does to the surface]
LENS:      [shot size + angle + focal length + aperture]
MATERIAL:  [the two surfaces that must read as real, described by behaviour]
FRAME:     [what is in the corners, and what is deliberately out of focus]

SETTINGS:  endpoint [flux-2-pro-preview] · width [1536] · height [1024]

The labels are scaffolding for you, not syntax for the model. Send prose.

Here is the same template filled in, so you can see the density the model actually rewards:

An editorial environmental portrait of a ceramicist in her workshop, mid-fifties,
grey hair pinned back with a pencil, dry clay dust on her forearms. She stands at a
wheel three feet from a north-facing window; behind her, shelves of unfired bisque
recede into shadow. Cool overcast daylight from camera left rakes across her face,
leaving a soft shadow under the jaw and catching the fine hairs at her temple. Shot
at 85mm, f/2, eye level, medium close-up. Skin shows visible pores across the nose
and cheek, with a slight sheen on the forehead; the linen apron has a slubby weave
and a worn crease where it folds at the hip. The wheel edge is in the bottom corner,
softly out of focus.

That prompt names one capture context, one light source with a direction, one lens, and two materials by how they behave. It contains no quality boosters at all. The word "photorealistic" does not appear in it, which is the point.

Which FLUX model should you send it to?

FLUX.2, in one of four variants. As of August 27, 2026 the image family is FLUX.2; FLUX.1 Kontext and FLUX1.1 pro are still live but BFL's own Kontext pages now say "For new projects, we recommend FLUX.2." FLUX 3, announced in July 2026, is on the API as a video model with synchronized audio. Its image modality is described as landing later, so nothing you write today goes to FLUX 3 as a still.

Three of those rows decide things.

Only flex exposes the dials. Its schema publishes steps from 1 to 50 with a default of 50, and guidance from 1.5 to 10 with a default of 5.0. The spec's own description of guidance is the sentence to remember for this whole page: "High guidance scales improve prompt adherence at the cost of reduced realism." If a render is coming back looking slightly illustrated, guidance is the first thing to lower, and flex is the only variant that lets you. Pro, max and klein take a prompt and dimensions and nothing else.

Klein is the one you can actually own. Klein Base 4B is published under Apache 2.0 on Hugging Face, so commercial use is permitted. Klein Base 9B and FLUX.2 dev are under the FLUX Non-Commercial License, which is a research and personal-work licence, not a client-deliverable one. BFL notes the Base variants are open weights for local development and research and "are not offered on the public API". Read the licence file before you invoice anyone for output.

Klein needs the most from your prompt. BFL states plainly that FLUX.2 klein "does not include prompt upsampling", and its quick reference tells you to write in prose rather than keyword lists. A prompt that pro quietly padded out for you will come back thin on klein. Every preset below is written at klein density for that reason, which also means none of them depends on the upsampler behaving.

How do you set resolution and aspect ratio in FLUX.2?

With width and height in pixels. There is no aspect_ratio parameter on any FLUX.2 endpoint.

This trips up people arriving from FLUX1.1 pro ultra or Kontext, both of which do take aspect_ratio as a string, described in the spec as a ratio "between 21:9 and 9:21". FLUX 3's video endpoint takes one too, from an enum. FLUX.2 does not. Its schema carries width and height with a documented minimum of 64 each and no published maximum, and BFL's FLUX.2 pages describe output "up to 4MP".

Treat 4 MP as the ceiling. BFL's outpainting endpoint spells the same limit out as 4,194,304 pixels and states you are free to choose any aspect ratio within that budget, and its video pricing footnote defines one megapixel as 1,048,576 pixels rather than a round million. The practical version:

Pick the ratio before you write the prompt, not after. BFL's guide puts it well: mismatched ratios "force the model to either crop or pad the composition". A prompt describing a wide landscape sent at 1:1 will get you a landscape with its ends amputated, and you will blame the prompt.

Iterate at 1024 × 1024 and only spend 4 MP on the frame you are keeping. FLUX.2 is priced per megapixel, so the resolution you test at is a budget decision as much as a quality one.

Does FLUX.2 support a negative prompt?

No. Not on any endpoint, not under any name.

This is worth stating flatly because a lot of FLUX wrapper sites still render a negative-prompt box. Three checks, all against BFL's own material on August 27, 2026:

1. The prompting guide's Technical Parameters page says: "Most FLUX models do not support negative prompts."
2. The FLUX.2 pro and max quick reference lists, as its first bullet, "No negative prompts supported".
3. Searching BFL's complete OpenAPI spec for the string negative returns exactly two hits, and both are the outpainting endpoint explaining that "Negative values are allowed" for a pixel offset. No field named negative_prompt exists anywhere in the spec.

BFL's own reasoning for the design is the interesting part, and it generalises past FLUX: "Even when they can process them, AI models generally struggle with negation". Its worked example is that writing "a person without glasses" causes the model to focus on the glasses and often generate exactly what you were trying to avoid. The guide's replacement drill is three steps: identify the unwanted element, ask what would fill that space instead, then describe that positively. It ships a table of swaps, including "no people" becoming "empty", "deserted", "solitary", and "not dark" becoming "brightly lit", "sun-drenched".

For photorealism specifically, the useful swaps are the ones about surface. Instead of "not plastic-looking skin", write what real skin does: pores across the nose, a slight sheen on the forehead, fine hairs catching the light. Instead of "no CGI look", name the capture: available light, handheld, slight motion in the hands. You get more control from describing the thing you want than you ever got from a negative box.

Why does "8k, ultra detailed, masterpiece" do nothing here?

Because it is a habit from a different architecture, and FLUX.2 is not that architecture.

The tag-soup style came from the Stable Diffusion era, where prompts were comma-separated token piles and a booster like "masterpiece" or "trending on artstation" nudged the ranking toward images that happened to carry those captions. It was never a resolution instruction. It was a vibe lever, and on models that read grammar it mostly wastes attention.

BFL's guidance is direct about this. On realism cues: "Avoid stacking too many generic quality terms. One or two strong realism cues are usually enough." On prompt length: "Start short. Add only what changes the image. More words do not automatically mean better results." And the line worth pinning above the desk: "Specific detail helps. Filler hurts."

There is a second reason boosters are worse than useless on FLUX.2 in particular. Resolution is a parameter here, set by width and height. Writing "8k" into the prompt does not change the pixel count of anything; it just tells a model that reads sentences that you would like the picture to be described as 8k, which is a caption, not a render setting. Meanwhile FLUX.2 accepts prompts "up to 32K tokens", so you have no length pressure forcing you into shorthand. There is nothing to gain by compressing craft into adjectives.

Swap the boosters for the four things that do move the render: a named capture context, a light source with a direction and a quality, a lens with an aperture, and two materials described by behaviour. That is the entire method, and every preset below is built from those four.

What actually moves a FLUX render toward photographic?

Naming the photograph rather than asking for realism. Four levers, in the order they matter.

Capture context beats the word "photorealistic". "Photorealistic" is a category, and categories average. "A 2000s digicam snapshot with on-camera flash" is a set of physical consequences: harsh frontal light, falloff into the background, slight chromatic noise, a slightly wrong white balance. BFL's own style table lists exactly these era looks, from "shot on Kodak Portra 400, natural grain, organic colors" to "early digital camera, slight noise, flash photography, candid, 2000s digicam style". Pick a capture, and realism arrives as a side effect.

Lighting carries the most weight per word. BFL's style page tells you to describe it the way a photographer would rather than settling for "good lighting", and its cheat sheet names the vocabulary the model responds to: golden hour, blue hour, overcast, Rembrandt, split lighting, chiaroscuro, rim light, practical lighting, diffused, harsh direct. Give each one a direction and a consequence. "Rembrandt lighting, key at 45 degrees" is fine; "Rembrandt lighting, key at 45 degrees, a small triangle of light under the far eye and a hard shadow line down the cheek" is better, because the second version tells the model what the light does when it lands. If you want the full vocabulary, 40 lighting terms for AI image prompts is the reference.

Lens language is a composition instruction. BFL's cheat sheet maps focal lengths to behaviour: 24mm shows more of the scene, 35mm is documentary, 50mm neutral, 85mm portrait-ideal with slight compression, 135mm and up strong compression. Aperture maps to depth: f/1.4 to f/2.8 blurs the background, f/8 to f/16 keeps everything sharp. Two warnings from experience. A wide focal length plus a tight framing request pulls in opposite directions, and the model will usually give you the wider frame. And BFL's guide notes prompt order matters here: if the model keeps pulling too far back, put the subject first and the environment later in the sentence.

Materials are where the render is won or lost. This is the part most prompts skip, and it is the reason FLUX.2's photorealism is worth using. Describe surfaces by behaviour rather than by adjective:

• Skin. Pores visible across the nose and cheek, a slight sheen on the forehead and nose bridge, fine vellus hair catching backlight at the jawline, a faint flush at the ears, one asymmetry such as a scar or a mole.
• Fabric. Weave first, then wear. Slubby linen with a broken crease at the hip, brushed cotton with pilling at the cuff, worsted wool with a dry matte finish and a sharp edge at the lapel, silk with a specular highlight that moves along the fold rather than sitting on it.
• Metal. Anisotropic brushed steel where the highlight smears along the grain, cast aluminium with a slightly grainy diffuse surface, brass with warm reflections and a dark tarnish in the recesses, chrome that reflects the room rather than glowing.
• Glass and liquid. Refraction bending what is behind it, a bright caustic on the surface below, a meniscus at the rim, condensation with droplets of two different sizes, one running.
• Wood, stone, ceramic. Grain that follows the shape of the object, a chipped glaze at the rim, dust settled in the recesses rather than evenly across the surface.

BFL's own photorealism page arrives at the same conclusion from the other side: prompt the model "as if a real photo is being captured in the moment", using photography language and explicitly asking for real texture, listing "pores, wrinkles, fabric wear, imperfections". Imperfection is a texture instruction, not a stylistic one.

25 copy-paste photorealism presets

Each preset is complete and parameterised. Square-bracketed slots are yours to fill; everything else is intended to survive as written. Every one carries a suggested endpoint and dimensions on the last line. They are written at klein density, so they work unchanged on pro, max and flex, with or without the upsampler.

Portrait

PORTRAIT 01 · Window light, available light only
An unposed environmental portrait of [subject: age, build, hair, one worn detail]
standing [distance] from a large north-facing window in [room]. Cool overcast
daylight from camera left wraps the face and falls off into the room behind. Visible
pores across the nose and cheek, a slight sheen on the forehead, fine hairs at the
temple catching the light. Shot at 85mm, f/2, eye level, medium close-up. Background
falls to soft grey-green at two stops under. Natural skin tone, no retouching.
SEND: flux-2-pro-preview · 1664 × 2496

PORTRAIT 02 · Rembrandt studio key
A studio portrait of [subject] against a mottled grey canvas backdrop. Single softbox
key at 45 degrees camera right and slightly above eye line, a small triangle of light
under the far eye, a hard shadow line down the cheek, and a white bounce card lifting
the shadow side by one stop. Shot at 105mm, f/4, chest-up. Skin texture retained,
pores and a faint razor line visible, catchlight square in both eyes.
SEND: flux-2-pro-preview · 1770 × 2360

PORTRAIT 03 · Backlit golden hour
[Subject] photographed outdoors twenty minutes before sunset, sun behind them at
frame right, rimming the hair and the edge of the shoulder in warm orange. Face lit
only by a dull bounce from the pavement, so it sits two stops under the rim. Slight
lens flare across the upper right. Shot at 135mm, f/2, waist-up, slight low angle.
Dust and pollen visible in the backlight. Warm highlights, cool shadows.
SEND: flux-2-max · 1664 × 2496

PORTRAIT 04 · 2000s digicam snapshot
A candid snapshot of [subject] at [event], taken on an early digital compact with
on-camera flash. Harsh frontal light, hard shadow on the wall behind, background
falling to near black. Slightly wrong white balance toward green, visible chromatic
noise in the shadows, mild over-sharpening at the edges. Off-centre framing, subject
mid-blink or mid-laugh. 2000s digicam style, candid, unedited.
SEND: flux-2-klein-9b-preview · 1600 × 1200

Product

PRODUCT 01 · Studio packshot, three-point
[Product] centred on a polished concrete surface in a studio. Three-point softbox
setup: broad key from upper left, fill from the right at half power, and a strip
light behind for edge separation. Soft diffused highlights, no harsh shadows, one
soft contact shadow directly under the object. Shot at 85mm, f/8, slightly above eye
level, medium shot. [Material: e.g. matte anodised aluminium with fine machining
lines] and [material 2] both legible at full sharpness. Commercial product photography.
SEND: flux-2-pro-preview · 2048 × 2048

PRODUCT 02 · Dark hero with practical rim
[Product] on a dark slate slab against a near-black background. Single hard key from
upper right at a steep angle, plus a narrow practical rim from behind camera left
tracing the silhouette. Deep shadows retain detail. Shot at 100mm macro, f/5.6, low
angle, three-quarter view. [Material] shows a specular highlight that smears along
the brushed grain rather than sitting as a dot. Fine dust visible on the slate.
SEND: flux-2-max · 2496 × 1664

PRODUCT 03 · Lifestyle in context
[Product] in use on [surface] in [real room], morning daylight through a side window,
sheer curtain diffusing it. Subject's hands enter frame from the right, skin texture
and a chipped nail visible. Shot at 35mm, f/2.8, table height, wide-ish framing that
keeps the room readable. Real clutter in the background at two stops under: a mug
ring, a stack of post, a cable. Nothing styled to perfection.
SEND: flux-2-pro-preview · 2360 × 1770

PRODUCT 04 · Macro material study
Extreme macro of [product detail: a watch crown, a stitched seam, a bottle thread] at
1:1 magnification. Raking side light from camera left at a shallow angle to exaggerate
surface relief. Shot at 100mm macro, f/8, focus-stacked so the whole detail is sharp
front to back. [Material] shows tool marks, a micro-scratch, and dust settled in the
recesses rather than evenly. Background falls to a smooth dark gradient.
SEND: flux-2-flex · 2048 × 2048 · steps 50 · guidance 3.5

Food

FOOD 01 · Overhead flat lay, daylight
Overhead shot of [dish] on [plate or board], shot from directly above. Soft daylight
from a window at the top of frame, a single black card at the bottom deepening the
shadows. Shot at 50mm, f/5.6, bird's-eye view. Crumbs, a smear of sauce off the edge,
one fork resting off-centre. [Ingredient] shows moisture and a slight sheen; [herb]
shows a curled edge where it has started to dry. Composed using rule of thirds.
SEND: flux-2-pro-preview · 2048 × 2048

FOOD 02 · Side-lit hero plate
[Dish] on a matte ceramic plate on a dark wood table, lit from directly camera left by
a large diffused source, with a bounce card just out of frame on the right. Shot at
85mm, f/2.8, plate height, three-quarter angle, front third of the dish sharp and the
back falling soft. Glaze catches a long specular highlight along the rim. Steam rising
and lit from behind. Warm neutral palette, no colour cast.
SEND: flux-2-max · 2496 × 1664

FOOD 03 · Motion and steam close-up
Close-up of [hot dish or drink] the moment [action: cream is poured, sauce is
spooned]. Backlit from camera rear-right so the steam reads bright against a dark
background. Shot at 100mm macro, f/4, 1/1000 second so the pour is frozen with a
crisp edge and a few discrete droplets. Condensation on the vessel with droplets of
two different sizes, one running. Cool background, warm subject.
SEND: flux-2-flex · 1664 × 2496 · steps 50 · guidance 4

Architecture

ARCHITECTURE 01 · Blue hour exterior
[Building type and style] photographed twelve minutes after sunset. Sky at deep blue
hour, interior lights warm and already on, the two colour temperatures balancing
against each other. Shot at 24mm tilt-shift, f/8, vertical lines kept perfectly
parallel, tripod height. Wet pavement in the foreground carrying reflections of the
windows. [Material: e.g. board-marked concrete] shows its grain and a rain stain
under the sill. No people, empty street.
SEND: flux-2-max · 3108 × 1332

ARCHITECTURE 02 · Daylight interior
Interior of [room type] photographed mid-morning, single large window at frame left
providing all the light, falloff across the floor toward the right. Shot at 24mm,
f/9, camera at 1.4 metres with vertical lines parallel. Real occupancy: a coat over a
chair, a book face-down, a slightly rumpled rug corner. [Material] on the floor and
[material 2] on the walls both legible. Natural white balance, no HDR halo.
SEND: flux-2-pro-preview · 2720 × 1530

ARCHITECTURE 03 · Detail and junction
Tight architectural detail of [junction: where the steel column meets the stone
plinth, where the glazing bead meets the frame]. Overcast light, no direct sun, so
every surface is described by its own texture rather than by shadow. Shot at 90mm
tilt-shift, f/11, square-on and centred. [Material] shows weathering, a sealant line
with a slight imperfection, and dirt collected in the shadow gap.
SEND: flux-2-flex · 2048 × 2048 · steps 50 · guidance 4.5

Landscape

LANDSCAPE 01 · Golden hour wide
[Landscape] photographed in the last thirty minutes of light, sun low at frame right
raking across the terrain and throwing long shadows toward the camera. Strong
foreground element: [rock, fence post, water edge] within two metres of the lens.
Shot at 20mm, f/11, hyperfocal, tripod at knee height. Haze building in the far
distance, three distinct depth planes. Natural colour, no graduated filter look.
SEND: flux-2-max · 3108 × 1332

LANDSCAPE 02 · Overcast minimal
[Landscape] under thick even overcast, no visible sun, no shadows. Flat light that
flattens the mid-tones and lets texture do the work. Shot at 70mm, f/8, eye level,
horizon on the lower third. Muted palette of [two colours]. A single small human-made
element far in the frame for scale. Fine drizzle visible against the darker mass.
Restrained contrast, slight cool cast.
SEND: flux-2-pro-preview · 2496 × 1664

LANDSCAPE 03 · Long exposure water
[Coast or river] photographed at a 30-second exposure at blue hour. Moving water
resolves into a soft continuous surface while [static element: rock, jetty, tree]
stays razor-sharp. Shot at 24mm, f/11, ISO 100, tripod, six-stop filter. Cloud
movement smeared into streaks converging toward the horizon. Wet rock in the
foreground with a specular sheen and visible barnacle texture. Cool, restrained.
SEND: flux-2-flex · 3108 × 1332 · steps 50 · guidance 3

Editorial

EDITORIAL 01 · Location fashion, full length
A full-length editorial fashion frame of [subject] in [garment] standing in [location].
Hard late-afternoon sun from camera right, sharp shadow of [object] falling across the
wall behind. Shot at 50mm, f/4, slight low angle, subject occupying the left third with
architectural negative space to the right. [Fabric] shows its weave and a real crease
where it folds at the hip; the hem moves in the wind. Confident stance, unstyled hair.
SEND: flux-2-max · 1770 × 2360

EDITORIAL 02 · Beauty close-up
A beauty close-up of [subject], face filling the frame from brow to chin. Large
diffused source directly above and slightly forward, plus a silver reflector below
throwing light back into the eyes. Shot at 100mm macro, f/5.6, dead square-on. Skin
kept fully textured: pores, a faint scar at [location], the fine down along the jaw
catching the top light. [Makeup or product] shows real edges and a slight sheen.
SEND: flux-2-pro-preview · 2048 × 2048

EDITORIAL 03 · Documentary reportage
A reportage frame of [subject] mid-[action] in [workplace]. Mixed lighting: cool
daylight through a high window and warm overhead fluorescent, both left uncorrected.
Shot at 35mm, f/2.8, handheld at chest height, subject slightly off-centre and one
foreground element cutting into frame. Slight motion blur in the hands, everything
else sharp. Grain consistent with ISO 1600. Nothing arranged for the camera.
SEND: flux-2-pro-preview · 2496 × 1664

Street

STREET 01 · 35mm daylight documentary
A candid street frame in [city, neighbourhood] at [hour]. Hard directional sun creating
a shape of light across the pavement that [subject] is about to walk into. Shot at
35mm, f/5.6, eye level, handheld, zone-focused. Layered depth: someone in the near
foreground cut by the frame edge, the subject at mid distance, signage behind. Grain
consistent with 400-speed film. Nobody looking at the camera.
SEND: flux-2-pro-preview · 2496 × 1664

STREET 02 · Night, practical light only
A night street frame in [city] lit entirely by practical sources: shopfront neon, a
sodium streetlamp, and headlights sweeping through frame left. No flash, no fill. Shot
at 35mm, f/1.8, handheld, ISO 3200, visible luminance noise in the shadows. Wet asphalt
carrying long colour reflections. [Subject] half-lit and half-lost in shadow, mid-stride.
Mixed white balance left uncorrected.
SEND: flux-2-max · 2496 × 1664

Still life

STILL LIFE 01 · Chiaroscuro tabletop
A still life of [objects] on a dark weathered table against a black background. Single
hard source from upper left at a steep angle, everything else falling to shadow, in the
manner of a Dutch old master. Shot at 90mm, f/11, slightly above table height,
focus-stacked. [Material] catches one bright specular edge; [second material] stays
matte and absorbs. Dust visible in the beam. Deep shadow retains detail.
SEND: flux-2-flex · 1770 × 2360 · steps 50 · guidance 4

STILL LIFE 02 · Glass and liquid
[Glass vessel] holding [liquid] on a [surface], backlit through a large diffused panel
with two black flags either side to draw dark edge lines down the glass. Shot at 100mm
macro, f/8, dead level with the liquid line. Refraction bends the background behind the
glass; a bright caustic falls on the surface below; a meniscus curves at the rim.
Condensation with droplets of two sizes, one running. No visible reflection of the studio.
SEND: flux-2-max · 1664 × 2496

STILL LIFE 03 · Fabric and texture study
A textile study of [fabric] draped over [form], filling the frame. Raking light from
camera left at a very shallow angle so the weave casts its own micro-shadows. Shot at
100mm macro, f/8, focus-stacked. The slub of the yarn, a pulled thread, and a worn
crease are all legible. [Second material] enters the corner for contrast. Neutral grey
background, no colour cast, exposure placing the fabric mid-grey.
SEND: flux-2-flex · 2048 × 2048 · steps 50 · guidance 4

One structured version

BFL documents structured JSON prompting for FLUX.2 and says the model "understands both formats equally well". Use it when a preset needs to be produced programmatically or when one element changes per SKU:

{
  "scene": "Professional studio product photography on a polished concrete surface",
  "subjects": [
    {
      "description": "[Product], [material], [finish]",
      "position": "Centre foreground",
      "color_palette": ["#1C1C1C", "#C9C4BC"]
    }
  ],
  "style": "Ultra-realistic product photography, commercial quality",
  "lighting": "Three-point softbox setup, soft diffused highlights, no harsh shadows",
  "mood": "Clean, professional, minimalist",
  "background": "Polished concrete with studio backdrop, falling two stops under",
  "composition": "rule of thirds",
  "camera": {
    "angle": "slightly above eye level",
    "distance": "medium shot",
    "lens-mm": 85,
    "f-number": "f/8",
    "ISO": 100,
    "focus": "Sharp on the front face, falling soft at the rear edge"
  }
}

Hex codes belong in that palette array. BFL documents exact colour matching from hex values, which is the cleanest way to hold a brand colour steady across a set.

What still gives a FLUX render away?

Five things, in our experience running these presets. BFL does not publish a limitations list for FLUX.2 text-to-image, so treat this section as our own observation rather than vendor guidance.

Lighting that is too well-behaved. Generated light tends to be tidier than real light: no stray reflection off an unseen surface, no colour contamination from a wall two metres away, no lens flare where a real lens would have flared. Naming a mixed-temperature situation and leaving it uncorrected is the cheapest fix, which is why several presets above do exactly that.

Backgrounds that are competent and pointless. Real photographs have accidents in them. Generated ones have arrangements. Asking for specific clutter, an object cut by the frame edge, or a person out of focus at the wrong depth reintroduces the mess that reads as documentary.

Repeated micro-texture. Look closely at brickwork, gravel, crowds and knitwear and you will often find the same unit tiling. It is most visible at 4 MP and least visible when the surface is partly in shadow or partly out of focus, so an aperture that puts some of it soft is doing double duty.

Hands, text and small hardware. FLUX.2 is much better here than earlier models and BFL leans on typography as a strength, but a zip pull, a watch crown, or a five-word sign is still where a render breaks. Check those regions at full resolution before anything ships. If your text is coming back scrambled across models, why is the text in my AI image garbled covers the general case.

Faces that are slightly too symmetrical. Real faces are asymmetric, and the asymmetry is what makes them read as a person rather than a rendering. Naming one specific asymmetry, a scar, a crooked incisor, one eyelid heavier than the other, is worth more than any amount of "hyper-realistic detail".

And the honest one about us. Prompt Architects does not generate images. We generate the prompt. You paste it into the BFL playground, an API call, or whichever host you use, and they render the pixels. What we save you is the thirty minutes of rewriting the same packshot brief for the eleventh SKU, and the slow drift where three people on a team each keep a slightly different version of it in a notes app.

The image prompt tools live in the web app and in the Chrome extension, and Global Variables fill the bracketed slots in the presets above, so one portrait preset covers a whole shoot list. Image prompt generation is available from the Pro plan upward. The free plan publishes 5 prompt enhancements per day, forever, which is enough to test whether the workflow suits you before any money changes hands, and current pricing including the launch discount is on the pricing page. If Google's model is where you actually work, the equivalent tool for it is our free Nano Banana Pro prompt generator.