Photorealism in FLUX.2 (by Black Forest Labs) comes down to three habits: lead with the right words, brief the camera like a photographer, and describe real texture instead of banning fake texture. FLUX.2 reads full descriptive sentences — not Midjourney tag-soup — so you write it a scene, not a pile of flags. This guide gives you the formula, the exact camera and lighting language, the guidance-scale settings that matter, and ten complete prompts you can paste today.
If you just want ready-made prompts, start with the 40 best Flux prompts roundup. For a themed set, see 26 Flux prompts for photorealistic images, and keep the Flux prompt cheat sheet open while you build your own.
The photoreal formula
FLUX.2's framework is Subject + Action + Style + Context, and word order matters: the model weights whatever comes first most heavily. Put the main subject at the front, then the key action, then the critical style, then the essential context, then any secondary detail. Write it as one descriptive paragraph, not a list.
- Subject — who or what, described concretely (age, build, clothing, material, expression).
- Action — what the subject is doing, even if subtle ("mid-stride", "pouring espresso").
- Style — photorealistic, plus the camera, lens, and film-stock language that carries the look.
- Context — setting, time of day, weather, surfaces, and the lighting that ties it together.
Length: keep it in the 30–80 word sweet spot. Ten to thirty words is fine for a quick concept; 80+ only pays off on complex multi-subject scenes. Enough to specify subject, camera, lighting, and context — not so much that the model loses the thread.
Skeleton to copy:
[Subject, described concretely] [doing a specific action], photorealistic,
shot on [camera body and lens], [film stock or colour look], lit by
[lighting setup], in [setting with real surfaces and time of day],
[texture cues: skin pores / fabric weave], shallow depth of field.
3:2 landscape, ~1536px, 2MP.The same skeleton, filled in:
A weathered fisherman in his sixties in a knitted grey sweater, looking just
off-camera, photorealistic, shot on Sony A7R V, 85mm f/1.4, Kodak Portra 400
colour, lit by soft golden-hour backlight with a gentle rim, on a misty
harbour wall at dawn, visible skin pores and salt-crusted stubble, shallow
depth of field. 4:5 portrait, ~1536px, 2MP.Why it works: the subject and action lead, "photorealistic" plus a real body and lens set the style, honest dawn light and texture cues sell the realism, and every element is specific enough to read like a shot a photographer could take.
Camera, lens & film language
Real gear names are the single strongest photoreal lever in FLUX.2. A camera body, a lens, and a film stock each carry a whole look — focal compression, depth of field, colour palette, and grain — that generic words never trigger. Write them the way a photographer would.
- Bodies & lenses: "Shot on Sony A7R V, 85mm f/1.4", "Canon EOS R5, 35mm f/1.8", "Hasselblad medium format, 80mm". The focal length sets the framing; the aperture sets the background blur.
- Film stocks: "Kodak Portra 400" (warm, soft skin), "Fujifilm Pro 400H" (cool pastels), "Cinestill 800T" (tungsten night look), "Ilford HP5" (classic black-and-white grain).
- Depth of field: "shallow depth of field, creamy bokeh" isolates a subject; "deep depth of field, sharp front to back" suits landscapes and interiors.
Specific beats generic. "Professional photo" is weak — the model reads it as vague. Name the tool instead and the realism follows. FLUX.2 also honours hex colour codes (drop #FF6B35 in for an exact brand hue) and renders in-image text well when you wrap the exact words in "quotes."
Lighting & texture cues
Lighting sets mood faster than any other factor, and texture cues are what push a face from render to photograph. Name a real lighting setup, then add the small physical details a camera would actually capture.
Lighting to name: golden-hour backlight, soft overcast diffusion, Rembrandt lighting, three-point softbox, hard direct sunlight, neon practical lights on wet asphalt, candlelight. One clear setup beats three fighting each other.
Texture cues (describe what you want, never what you don't):
- Skin: "visible skin pores, fine flyaway hairs, subtle under-eye shadows, natural blemishes, no retouching." This is how you kill the plastic look — with positive detail, not a negative prompt.
- Fabric: "visible weave in the linen, frayed denim seams, soft wool pilling, worn leather creases."
- Surfaces: "condensation beads, dust motes in the light, fingerprints on glass, scuffed matte metal."
Because FLUX.2 has no negative prompts, every one of these is phrased as something to include. Want fewer artefacts? Add the real detail you want the model to render in their place. See the photorealism collection for these cues in finished prompts.
Guidance, steps & resolution
A great prompt still needs the right settings. Three levers matter for realism: guidance scale, steps, and resolution.
- Guidance scale: default 2.5 gives natural, photographic results. Raise to 5–8 when you need FLUX.2 to follow the prompt literally — product, technical, or brand-exact work. Lower it for more creative freedom.
- Steps: default 30. Use 15–20 for fast drafts and up to 100 for maximum detail. Only the [flex] tier exposes the steps and guidance sliders; on [pro] and [max] they are fixed to optimal internal values, and on [klein] they have no effect.
- Resolution: FLUX.2 goes up to 4MP (2048×2048), but 2MP (~1440–1536px) is recommended for most work and renders faster. Output dimensions must be multiples of 16. Set aspect ratio with width and height: 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, 9:16.
Pick your tier by need: [pro] for top quality and text, [max] for highest fidelity, [flex] for manual control, [klein] for sub-second drafts, and [dev] open weights to run locally in ComfyUI.
10 example prompts
Ten complete photoreal prompts that assemble the formula above. Paste any into the bfl.ai playground, the FLUX API, or a partner platform and tweak the specifics. For more, browse the best Flux prompts.
1. Portrait
A confident woman in her thirties with short natural curls and a tailored
linen blazer, looking directly at camera, photorealistic, shot on Sony A7R V,
85mm f/1.4, Kodak Portra 400 colour, lit by soft Rembrandt window light in a
sunlit loft, visible skin pores and fine flyaway hairs, shallow depth of field.
4:5 portrait, ~1536px, 2MP.Why it works: pore-level texture plus a real portrait lens stops the face reading as plastic.
2. Food
A rustic sourdough loaf torn open to show an airy crumb, steam still rising,
photorealistic, shot on Canon EOS R5, 100mm macro f/2.8, lit by soft overcast
window light from the left, on a flour-dusted oak board with a linen cloth,
visible crust blistering and flour grains, shallow depth of field.
3:2 landscape, ~1536px, 2MP.Why it works: macro lens plus crumb-level texture cues make the bread look freshly baked.
3. Product
A matte-black stainless steel water bottle standing three-quarter view,
photorealistic, shot on Sony A7R V, 90mm macro, lit by a three-point softbox
setup with a crisp rim light, on a seamless #F2F0EB backdrop with a soft
reflection, condensation beads and brushed-metal grain, deep depth of field.
1:1 square, ~1536px, 2MP. Guidance scale 6 for literal, e-commerce accuracy.Why it works: a raised guidance scale and hex backdrop keep the product true for a listing.
4. Street
A commuter in a wool overcoat mid-stride crossing a rain-slicked intersection
at dusk, photorealistic, shot on Fujifilm X100V, 35mm, Cinestill 800T look, lit
by magenta and cyan neon reflecting off wet asphalt, blurred traffic behind,
grain and motion in the background, deep depth of field.
3:2 landscape, ~1536px, 2MP.Why it works: the tungsten film look and neon practicals give an authentic night-street feel.
5. Wildlife
A red fox pausing alert in tall frosted grass at first light, photorealistic,
shot on Nikon Z9, 400mm f/2.8 telephoto, lit by low golden-hour sun from behind
rimming the fur, misty meadow bokeh in the distance, visible individual guard
hairs and breath vapour, very shallow depth of field.
3:2 landscape, ~1536px, 2MP.Why it works: the long telephoto and hair-level detail read as a real wildlife capture.
6. Interior
A sunlit Scandinavian living room with a pale oak floor and a linen sofa,
photorealistic, shot on Canon EOS R5, 24mm f/4, lit by soft morning light
raking through sheer curtains, dust motes drifting in the beams, visible wood
grain and fabric weave, deep depth of field, everything sharp front to back.
16:9 wide, ~1536px, 2MP.Why it works: a wide lens with deep focus and dust motes gives an honest architectural shot.
7. Golden-hour landscape
A winding mountain river valley at golden hour, mist pooling between pine
ridges, photorealistic, shot on Sony A7R V, 24mm f/8, lit by warm low sun
casting long shadows across the slopes, layered atmospheric haze, crisp
foreground rocks and distant peaks, deep depth of field.
3:2 landscape, ~1536px, 2MP.Why it works: atmospheric haze plus long golden shadows create real depth and scale.
8. Fashion
A model in an oversized cream trench coat leaning against a concrete wall,
editorial pose, photorealistic, shot on Hasselblad medium format, 80mm, lit by
hard direct afternoon sun with sharp shadows, on a minimalist urban street,
visible fabric weave and coat creases, natural skin texture, shallow depth of
field. 2:3 portrait, ~1536px, 2MP.Why it works: medium-format language and hard sun deliver a high-end editorial look.
9. Candid
Two friends laughing mid-conversation over coffee at a small cafe table,
unposed candid moment, photorealistic, shot on Fujifilm X-T5, 50mm f/1.8,
Fujifilm Pro 400H colour, lit by soft window daylight, warm bokeh of the
cafe behind them, natural skin texture and genuine expressions, shallow depth
of field. 3:2 landscape, ~1536px, 2MP.Why it works: "unposed" plus natural daylight and real expressions avoid the stiff studio feel.
10. Film-look
A young man in a denim jacket leaning on a vintage car at dusk, photorealistic
1970s film aesthetic, shot on 35mm, Kodak Portra 800 with visible grain and
slight halation, lit by warm fading sunlight and a soft lens flare, faded
muted colours, natural skin texture, shallow depth of field.
3:2 landscape, ~1536px, 2MP.Why it works: named film stock, grain, and halation reproduce a genuine analog-camera look.
Mistakes to avoid
Most FLUX.2 realism failures come from prompting it like an older model. Fix these five and outputs improve immediately.
- Using
--flags. FLUX.2 takes no Midjourney parameters — no--ar,--style, or--v. State aspect ratio and settings in plain words or via the API fields. - Writing negative prompts. There are none. "No plastic skin" does nothing; "visible skin pores and natural texture" does the work. Always describe what you want.
- Vague "professional photo". It reads as weak and generic. Name a real body, lens, and film stock instead — that specificity is what triggers the photoreal look.
- Overstuffing. Cramming 150 words of adjectives dilutes the important elements. Stay in the 30–80 word sweet spot and cut anything that isn't pulling weight.
- Ignoring word order. Burying your subject at the end weakens it. FLUX.2 weights early words most, so lead with the subject, then action, then style, then context.
Keep the Flux prompt cheat sheet handy for the tables of tiers, lenses, and lighting while you write.
Frequently Asked Questions
What is the best prompt structure for photorealism in FLUX.2?
Use Subject + Action + Style + Context, and lead with the most important element. Word order matters in FLUX.2 — put the main subject first, then the key action, then the critical style (photorealistic, camera language), then the essential context, then secondary details. Write it as one descriptive paragraph in the 30–80 word sweet spot rather than a list of tags.
What guidance scale should I use for realistic FLUX.2 images?
The default guidance scale is 2.5, which balances prompt adherence with natural, photographic results. Raise it to 5–8 when you need FLUX.2 to follow the prompt more literally, such as product or technical shots. Note that only the [flex] tier exposes the guidance and steps sliders — on [pro] and [max] these are fixed to optimal internal values, and on [klein] they have no effect.
Should I use negative prompts in FLUX.2?
No. FLUX.2 does not support negative prompts. Instead of listing what you don't want, describe what you do want. If you keep getting plastic-looking skin, add positive texture cues like "visible skin pores, fine facial hair, natural blemishes" rather than writing "no plastic skin." Describing the target directly is how the model was trained to work.
How long should a FLUX.2 photorealism prompt be?
The sweet spot is 30–80 words. Short prompts of 10–30 words are fine for quick concepts, and long prompts of 80+ words suit complex multi-subject scenes, but most photoreal work lands best in the medium range. Enough words to specify subject, camera, lighting, and context — not so many that the model loses the thread.
What resolution and dimensions work best in FLUX.2?
FLUX.2 renders up to 4MP (for example 2048×2048), but 2MP — roughly 1440–1536px on the long edge — is recommended for most work and is faster. Output dimensions must be multiples of 16. Set your aspect ratio with width and height; common choices are 1:1, 3:2, 2:3, 4:3, 3:4, 16:9, and 9:16.
How do I stop FLUX.2 faces from looking like plastic?
Add pore-level texture cues and real camera language. Phrases like "visible skin pores, natural skin texture, fine flyaway hairs, subtle under-eye shadows" plus a real lens ("shot on Sony A7R V, 85mm f/1.4") and honest lighting push FLUX.2 toward a photograph rather than a render. Avoid over-smooth style words like "beautiful" or "flawless," which nudge it back toward CGI.
Does FLUX.2 understand real camera and film-stock names?
Yes. FLUX.2 responds well to real camera bodies, lenses, and film stocks — for example "Shot on Sony A7R V, 85mm f/1.4" or "Kodak Portra 400." These carry a whole look: focal compression, depth of field, colour palette, and grain. Specific gear language beats vague terms like "professional photo," which the model reads as weak and generic.
Which FLUX.2 tier is best for photorealism?
Use FLUX.2 [pro] for top quality and the best text rendering, or [max] for the highest fidelity. Choose [flex] when you want manual control over steps and guidance, and [klein] for fast, sub-second drafts where those sliders don't apply. The open-weight [dev] tier runs locally in ComfyUI for offline work.