Tips & Tricks8 min read
Structured Prompt Templates: Six Slot Frameworks That Work
Structured prompts see 8× the median engagement of prose at the same length in our corpus. Here are six copy-ready slot templates and why naming the slots is what does the work.
On this page
- Why these templates exist
- Why named slots work
- Do you have to write valid JSON?
- Template 1: Portrait
- Template 2: Product shot
- Template 3: Poster with text
- Template 4: Multi-panel character sheet
- Template 5: Scene and story frame
- Template 6: Style transfer from a reference image
- Three ways people break these templates
- FAQ
- Should I write AI image prompts in JSON?
- Do AI models actually parse JSON prompts?
- Can I put a negative prompt inside the template?
- Which model handles structured prompts best?
- Where can I see more structured prompts?
Why these templates exist
13% of the 9,599 published prompts in our gallery are written as structured slots rather than prose. In the 80–149 word band — same length, different structure — those prompts have a median of 141 views against 17 for prose, roughly 8×. That gap is the reason this page exists.
Below are six complete templates you can copy: portrait, product, poster with text, multi-panel character sheet, scene, and style transfer. Each one is filled in with working values, so you can paste it, swap the parts that describe your subject, and run it. The corpus behind the numbers is documented in what 9,599 real prompts actually contain.
13%
of prompts are written as structured slots
1,249 items; the other 8,350 are prose
141 vs 17
median views, structured vs prose at 80–149 words
the length band where the gap is widest
100–145
words per template once filled in
inside the 80–149 band; corpus median prompt length is 83 words
Methodology, and what these numbers are not
Counts are exact, over 9,599 published prompts longer than 20 characters as of August 2026. Engagement figures are on-site view counts, which are also affected by how long an item has been published and where the feed surfaced it — correlation, not proof of causation. What we can say with confidence is that the structured/prose gap survives controlling for length: 141 vs 17 at 80–149 words, 67 vs 32 at 150–299, 98 vs 33 above 300.
Why named slots work
The mechanism is not magic and it is not JSON syntax. It is this: a named slot makes an unfilled slot visible.
Prose lets you write a beautiful sentence that never mentions where the light comes from. You read it back, it sounds finished, and the model quietly invents the lighting, the lens, the background and the mood. A template with a lighting key cannot be read back as finished while that key is empty. You are not being made more creative — you are being stopped from shipping a prompt with holes in it.
✕ Bad
A moody portrait of a woman in a cafe, beautiful lighting, cinematic, highly detailed, masterpiece✓ Good
subject: woman, late 20s | setting: corner table, empty cafe, mid-morning | lighting: window light from camera left, hard edge, no fill | camera: 50mm, f/2.8, eye level | mood: waiting, not sad | style: editorial photograph, fine grainThe bad version has four slots unfilled and does not look unfinished. That is exactly the trap.
90s Grunge Rose Flash Portrait
Nano Banana ProPrompt▾
{"image_analysis_prompt":{"subject":{"description":"Young woman with fair skin and shoulder-length, dark, messy wavy hair.","hair_styling":"Hair is tousled and hangs over the face, partially obscuring the left eye.","expression":"Alluring and edgy; mouth slightly open with teeth visible."},"facial_details":{"eyes":"Visible right eye is light-colored (blue/green) with natural eyelashes.","nose":"Small nose with a silver horseshoe septum piercing.","mouth":"Dark red lipstick, slightly glossy. Lips are parted.","markings":"A distinct, bright red lipstick kiss mark (imprint of lips) on the right cheek."},"props_and_action":{"object":"A fresh, deep red rose with green leaves and stem.","interaction":"The subject is holding the rose directly against her mouth, with the petals grazing her lips and teeth, suggesting a biting or kissing motion."},"attire":{"clothing":"A loose, light pink button-down shirt or robe.","accessories":"A thin black bra strap or tank top strap is visible on the shoulder."},"technical_style":{"lighting":"Direct flash photography style, high contrast with slightly overexposed skin highlights (typical of 90s grunge or indie sleaze aesthetic).","color_palette":"Dominant reds (rose, lipstick, kiss mark), dark browns (hair), and soft pinks (clothing/skin).","resolution":"4k, HD, hyper-realistic textures.","focus":"Sharp focus on the face and rose, with highly detailed skin texture and hair strands."},"full_prompt_string":"A highly detailed 4k HD portrait of a young woman with messy dark hair covering one eye, silver septum piercing, and dark red lipstick. A bright red lipstick kiss mark is stamped on her cheek. She is holding a red rose to her mouth, biting the petals. She is wearing a pink shirt with a visible black strap. The aesthetic is 90s grunge flash photography, realistic texture, edgy atmosphere, sharp focus, high quality."}}Do you have to write valid JSON?
No. Strict JSON syntax is not the active ingredient, and prompts in the corpus prove it — plenty of the highest-effort ones use numbered headings or pipe-separated labels rather than braces.
All three of these work:
Format A — JSON
{
"subject": "...",
"lighting": "...",
"camera": "..."
}
Format B — labelled lines
Subject: ...
Lighting: ...
Camera: ...
Format C — numbered sections
I. Subject
1. Description: ...
2. Pose: ...
II. Light
1. Key: ...
Pick whichever you will actually keep filling in. JSON has one practical advantage worth knowing: nesting. When a slot needs its own sub-slots — nine panels, three text blocks, a wardrobe with five items — braces keep the structure legible where labelled lines start to sprawl.
Portrait: Wind-Swept Dark Hair, Eyes Concealing a Lonely Shadow
Nano Banana ProPrompt▾
I. Subject & Action 1. Subject: young woman, cool/aloof temperament, half-body close-up, body turned to the right, looking back over her shoulder directly at the camera 2. Expression: eyes sharp and restrained, slightly detached and melancholic, lips gently pressed 3. Hair: jet-black long hair, voluminous and tousled, strong wind blowing, strands dramatically lifted and sweeping across the cheek 4. Makeup: warm brown eyeshadow with slightly reddened eye area, defined eyelashes, matte red lips, translucent skin texture 5. Clothing: dark thin-strap dress or top, shoulders and clavicle exposed, outer side with dark sheer fabric or an outer layer slipping down creating a wrapped shadowed area II. Composition & Lens 1. Orientation: vertical 2. Framing: half-body close-up above the chest, three-quarter profile 3. Positioning: subject slightly to the mid-right of the frame, small headroom, left side reserved for light source and diffusion area 4. Angle: eye-level or slight low-angle, camera distance close but without exaggerated distortion 5. Focus: focal point on the eyes and face, strong background bokeh, overall slight softening and haze III. Lighting & Atmosphere 1. Key light: strong backlight from left-rear creating bright rim/rim light on hair and shoulder 2. Fill light: very weak frontal soft fill, only boosting facial detail, shadows retained 3. Light quality: diffused soft light, pronounced veiled halo/bloom, airy feel with slight lens flare 4. Mood: dreamy, mysterious, quiet, cinematic, with a touch of coolness and loneliness IV. Color & Texture 1. Color grading: background cool gray-blue, skin tones warm, creating cold-warm contrast 2. Saturation: overall low saturation / desaturated 3. Tonal range: relatively high contrast, crushed shadows, highlights soft and not blown 4. Texture: skin delicate and natural, slight grain and matte finish, edge highlights with glow/bloom V. Background & Environmental Elements 1. Background: gradient space from dark gray to black, no obvious scene information 2. Light particles: scattered point light specks and flickering particles around the background and hair tips, like dust reflections or micro light particles 3. Motion: wind-driven flying hair, light particles floating and layered for depth VI. Style & Post-processing 1. Style: beautiful realism, dreamy soft haze, film poster aesthetic 2. Post-processing: enhance rim and hair highlights, increase haze and translucence of light, slight fade/desaturation, vignette to focus the frame 3. Sharpness: eyes and lips relatively sharp, other areas softened with smooth transitions, overall translucent but not overly sharp VII. Negative Prompts / What to Avoid 1. Subject errors: no facial deformation, no misaligned features, no double faces, no extra limbs, no abnormal fingers, no unnatural plastic skin 2. Lighting errors: no hard frontal light, no strong blown white highlights, no overly bright background that competes with the subject 3. Style deviations: no cartoonization, no heavy painterly/comic thick-paint look, no exaggerated neon colors, no over-sharpening or over-smoothing
Template 1: Portrait
{
"subject": "woman, late 20s, shoulder-length black hair, minimal makeup, small mole under left eye",
"wardrobe": "charcoal ribbed turtleneck, thin gold chain, no other jewellery",
"pose": "seated, turned three-quarters to camera, looking back over her shoulder",
"expression": "calm, unposed, mouth relaxed, not smiling",
"setting": "bare plaster wall, indoor, mid-morning",
"lighting": "window light from camera left, soft falloff, faint warm bounce filling the shadow side",
"camera": "85mm, f/2.0, eye level, waist-up framing",
"focus": "sharp on the near eye, background fully dissolved",
"color": "muted neutrals, slightly warm skin tones, no colour cast on the wall",
"style": "editorial portrait photography, fine film grain, no retouching gloss"
}
Swap subject, wardrobe and setting. Leave focus and lighting structured the way they are — direction plus quality plus fill is the pattern that keeps working. The camera line follows the frequency data on lenses and apertures in the camera language article.
Gentle Melancholy of Backlit Hair Tips
Nano Banana ProPrompt▾
{
"subject": {
"description": "Young East Asian woman, positioned in profile turning to look back at the viewer.",
"pose": "Looking over shoulder, slight head tilt, body angled away from camera.",
"expression": "Soft, neutral to slightly melancholic, contemplative gaze, lips slightly parted."
},
"clothing": {
"upper_body": "Dark, structured jacket or blazer, possibly pinstriped or textured fabric.",
"visibility": "Mostly obscured by shadow and foreground elements."
},
"hair": {
"color": "Dark brown to black.",
"style": "Long, loose, tousled waves, slightly messy texture.",
"lighting_interaction": "Strongly backlit, creating a glowing halo effect on loose strands (rim lighting)."
},
"face": {
"skin_tone": "Fair/pale with warm undertones.",
"features": "Soft facial structure, almond-shaped eyes, natural eyebrows.",
"makeup": "Minimal, 'no-makeup' look, natural lip color."
},
"accessories": {
"visible_items": "None clearly visible due to framing and lighting."
},
"environment": {
"setting": "Cluttered interior space, resembling a workshop, storage room, or old bookstore.",
"background_elements": "Wooden shelves stacked with indistinct objects, papers, boxes, and plastic packaging.",
"foreground_elements": "Blurred translucent objects (possibly glass, plastic sheets, or dust covers) creating layers and reflections."
},
"lighting": {
"source": "Natural sunlight streaming through a window on the left.",
"quality": "Golden hour, warm, diffuse but directional.",
"effects": "Strong backlighting/rim lighting on hair, volumetric dust motes dancing in the light, lens flares, soft glare, dramatic contrast between light and shadow."
},
"camera": {
"perspective": "Eye-level, shot through a foreground obstruction (glass or clutter).",
"focus": "Shallow depth of field (bokeh), sharp focus on eyes/face, blurred background and foreground.",
"lens_character": "Soft focus, film grain simulation, slight bloom effect."
},
"style": {
"aesthetic": "Cinematic, atmospheric, nostalgic, slice-of-life, ethereal.",
"mood": "Dreamy, intimate, quiet, sentimental.",
"visual_reference": "Film photography, Japanese photobook style."
}
}Template 2: Product shot
{
"product": "matte black ceramic coffee mug, 350ml, unbranded, slightly irregular hand-thrown rim",
"hero_treatment": "single unit, tilted 15 degrees, floating just above the surface",
"surface": "polished travertine slab, warm grey",
"background": "seamless mid-grey, gentle top-to-bottom gradient, no props",
"lighting": "large softbox upper left as key, black flag on the right to keep that edge dark, thin rim highlight along the handle",
"camera": "50mm, f/4, slightly above product height",
"shadow": "one soft contact shadow directly beneath the mug, no secondary shadows",
"detail": "unglazed clay texture visible on the foot, single specular highlight on the rim",
"color": "warm neutral palette, no colour cast on the white areas",
"output": "4:5 vertical, product fills 60 percent of frame height, even margins"
}
The two slots people leave out are shadow and output. Unspecified shadows are where AI product shots read as fake — usually two or three of them pointing in different directions. Unspecified framing is why the product arrives too small to use.
Metallic Pink Perfume Luxury Ad
Nano Banana ProPrompt▾
{
"resolution": "8K",
"aspect_ratio": "3:4",
"style": "high-end AI-edited studio photography",
"subject": {
"primary": "human hand holding a metallic pink perfume bottle",
"hand_details": {
"skin": "smooth, well-groomed",
"nails": "short to medium length, natural nude tone",
"accessory": "white pearl bracelet on wrist"
},
"bottle": {
"material": "metallic reflective",
"color": "deep metallic pink",
"shape": "cylindrical with rounded shoulders",
"nozzle": "gold spray nozzle, cap removed",
"charm": {
"material": "gold",
"shape": "small geometric star-like tag",
"attachment": "hanging from bottle neck via thin metal loop"
},
"visible_text": "MONTALE"
}
},
"action": {
"spray": {
"type": "fine atomized mist",
"direction": "left to right",
"texture": "soft, smoke-like, semi-transparent"
}
},
"elements_emerging_from_mist": {
"flowers": {
"pink_roses": "fully bloomed roses floating in air",
"rose_petals": "individual pink petals suspended mid-air",
"white_flowers": "small star-shaped white blossoms with thin petals"
},
"butterfly": {
"color": "pink",
"wings": "semi-translucent with visible veins",
"position": "right side of frame"
},
"particles": {
"type": "small glowing bokeh-like specks",
"colors": ["soft pink", "warm gold"]
}
},
"background": {
"color": "dark black to deep brown gradient",
"environment": "studio, no visible location"
},
"lighting": {
"type": "soft cinematic studio lighting",
"highlights": ["bottle curves", "gold nozzle", "mist", "flowers"],
"contrast": "high subject-to-background contrast"
},
"composition": {
"orientation": "vertical",
"subject_position": "slightly left of center",
"depth": "foreground hand, mid-air mist and flowers, dark background"
},
"quality": {
"sharpness": "ultra-sharp",
"detail_level": "extreme micro-detail",
"finish": "luxury advertising grade"
}
}Template 3: Poster with text
{
"format": "vertical poster, 2:3",
"topic": "weekend coffee tasting event",
"text_blocks": {
"headline": "SLOW POUR",
"subhead": "Single-origin tasting, Saturday 10am",
"footer": "Bench Coffee, 42 Mill Street"
},
"typography": "heavy grotesque sans for the headline, light sans for the subhead, headline all caps, tight letter spacing",
"layout": "headline in the top third, image band across the middle, footer on the last line with a generous bottom margin",
"image_band": "overhead shot of a pour-over cone mid-pour, steam visible, warm wood table",
"palette": "cream background, near-black type, one burnt orange accent",
"spelling": "render every string exactly as written, no invented words, no additional text anywhere in the frame",
"style": "swiss editorial poster, flat, printed feel, no drop shadows"
}
Two slots do the heavy lifting here. Putting the copy in its own text_blocks object separates what the words are from how they look — models garble text far more often when the two are tangled in one sentence. And an explicit spelling rule is the only reliable defence against the invented sixth word that appears in the corner.
Text and layout work skews strongly by model: 51.7% of GPT Image prompts in the corpus ask for text or typography, against 21.8% for Nano Banana and 7.9% for Midjourney. If your work is layout-heavy, start from the GPT Image prompt collection.
Template 4: Multi-panel character sheet
{
"format": "3x3 grid, nine equal panels, thin white gutters, delivered as one image",
"character": "same person in every panel: woman, 20, silver bob, round wire glasses, oversized navy varsity jacket, white tee",
"consistency_rule": "identical face, hairstyle and outfit in all nine panels; treat panel 1 as the reference for the other eight",
"panels": [
"1 front view, neutral expression",
"2 three-quarter view, slight smile",
"3 profile, looking left",
"4 laughing, head tilted back",
"5 arms crossed, mid-shot",
"6 hands in pockets, full body",
"7 sitting on the floor, knees up",
"8 back view, head turned to camera",
"9 close-up of hands holding a paper cup"
],
"lighting": "same soft studio key in every panel, identical colour temperature throughout",
"background": "flat pale grey, identical in all nine panels",
"camera": "50mm, f/4, eye level unless the panel says otherwise",
"style": "clean editorial photography, consistent grain across panels"
}
The consistency_rule slot is the one that matters. Nine panels of a similar person is the default failure, and naming panel 1 as the reference is what pins the other eight. Grid sizes, drift repair and what to do when the ninth panel wanders are covered in the multi-panel guide.
Cat-Eared Student Council Character Reference
GPT ImagePrompt▾
{
"type": "anime character reference sheet",
"style": "clean polished light novel / visual novel design sheet, white background with pale lavender ornamental border, high-detail cel shading, elegant school-themed layout, soft cool lighting, crisp line art",
"character": {
"name": "霧島 澪奈",
"role": "serious female class committee president with black cat traits",
"age": "high school student, second year",
"species traits": [
"2 black cat ears on top of head with soft inner fur",
"1 long black cat tail"
],
"appearance": {
"gender presentation": "female",
"build": "slender, refined, modestly curvy",
"height note": "tall for a schoolgirl",
"hair": {
"color": "deep blue-black",
"style": "very long straight hair with slight waves at the ends, center-parted bangs, long sidelocks, half-up back section tied with a ribbon bow"
},
"eyes": {
"color": "golden amber",
"shape": "sharp but elegant"
},
"expression base": "cool, intelligent, reserved"
},
"outfit": {
"count": 9,
"pieces": [
"navy school blazer with white piping",
"white dress shirt",
"large striped ribbon tie in navy and burgundy",
"school crest on blazer chest",
"red armband with white Japanese text indicating committee leader",
"blue plaid pleated skirt",
"black semi-sheer tights",
"black loafers",
"thin gold-framed glasses"
]
},
"props": {
"count": 1,
"items": [
"dark school folder or notebook with crest"
]
}
},
"layout": {
"format": "landscape character design board",
"sections": [
{
"title": "プロフィール",
"position": "left column",
"count": 1,
"labels": ["bio text block"]
},
{
"title": "SIDE / BACK",
"position": "upper center",
"count": 3,
"labels": ["front full body", "side full body", "back full body"]
},
{
"title": "表情バリエーション",
"position": "upper right",
"count": 8,
"labels": ["通常（真面目）", "微笑み", "注意・叱責", "驚き", "照れ（目をそらす）", "困り顔", "優しい笑顔", "猫っぽく甘える"]
},
{
"title": "ディテール・アクセサリー",
"position": "middle right left panel",
"count": 5,
"labels": ["委員長の腕章", "校章", "猫モチーフのヘアピン", "リボンタイ", "細フレーム眼鏡"]
},
{
"title": "制服ディテール",
"position": "middle right right panel",
"count": 3,
"labels": ["chest and tie close-up", "waist and blazer trim close-up", "sleeve cuff and buttons close-up"]
},
{
"title": "カラーパレット",
"position": "lower left",
"count": 5,
"labels": ["髪", "瞳", "制服", "アクセント", "肌"]
},
{
"title": "特徴的なポーズ・仕草",
"position": "lower center",
"count": 4,
"labels": ["指示を出す（委員長モード）", "資料を確認する", "考え事をする", "休み時間の素顔（少し気が抜けた瞬間）"]
},
{
"title": "その他設定・ポイント",
"position": "lower right",
"count": 2,
"labels": ["classroom vignette", "moonlit night vignette"]
}
]
},
"text blocks": {
"headline": "霧島 澪奈",
"subheadline": "学級委員長",
"quote": "みんなのために、きちんとしなきゃ。それが、私の役目だから。",
"body text": "dense Japanese profile text with school year, birthday, height, blood type, club or committee, personality notes, likes and dislikes"
},
"composition notes": [
"main full-body figure at left, holding glasses near face and carrying a folder",
"side and back character turnaround in the center",
"all portraits and close-ups neatly framed in pale lavender boxes with decorative corner motifs",
"facial close-ups cropped at shoulders",
"bottom-right vignette images show the character in a classroom and outdoors at night under a crescent moon",
"overall palette is navy, black, lavender, white, burgundy, muted gold, and soft skin tones"
]
}Template 5: Scene and story frame
{
"scene": "late-night laundromat, one row of machines running, empty street visible through the window",
"subject": "man in his 30s, rumpled work shirt, sitting on a moulded plastic chair, holding a paper cup",
"action": "looking out of the window, not at the machines",
"time": "01:20, no other customers",
"lighting": "hard overhead fluorescent tubes with a green cast, weak warm spill from a vending machine on the right",
"atmosphere": "still air, faint condensation on the window glass",
"camera": "35mm, f/2.8, eye level from across the room, subject small in frame",
"composition": "subject on the left third, the empty machines filling the right two thirds",
"color": "green-cyan fluorescents against one warm accent, desaturated overall",
"mood": "patient and unremarkable, not sad",
"style": "cinematic still, 35mm film grain, no lens flare"
}
Note action and mood as separate slots from subject. Prose usually collapses all three into one sentence, and the first thing lost is the action — which is what makes a frame read as a moment rather than a pose. The not sad half of the mood line matters too: naming what you do not want stops the model reaching for the nearest cliché.
Ice Blue Interrogation Femme Fatale
Nano Banana ProPrompt▾
{
"scene_description": {
"location": "Cold interrogation room with teal-blue brick tiled wall, cinematic still from 'Basic Instinct' (1992)",
"vibe": "Intense seductive power, classic film-noir tension, commanding presence",
"lighting": "Teal-blue ambient room light, strong directional key from left casting long shadows, rim light sculpting figure and legs, high contrast, dramatic neo-noir mood",
"camera_settings": "35mm anamorphic lens, f/2.0, shallow depth of field, cinematic film grain, subtle lens breathing, ultra-photorealistic 8k, 'Basic Instinct' color palette (cool teal shadows, warm skin highlights)"
},
"primary_subjects": [
{
"character": "attached celebrity lookalike",
"role": "Interrogation subject / ultimate femme fatale",
"action": "Seated in black metal chair, legs crossed seductively (right over left), right arm casually draped over chair back, left hand resting on thigh, upper body leaning back slightly, hips angled forward, piercing direct stare at camera, lips parted in confident half-smile",
"details": "Sleeveless white high-neck mini dress clinging to curves, bare toned legs with glossy sheen, white pointed high heels, long dark glossy hair with loose waves, smoky cat-eye makeup, bold red lips, gold bracelet on right wrist"
}
],
"environment_details": {
"crowd": "None – solitary cinematic intensity",
"architecture": "Teal-blue brick tiled wall, black metal chair, concrete floor",
"fidelity": "Hyper-realistic 8k, lifelike skin under cold light, fabric detail, hair shine, precise 'Basic Instinct' Catherine Tramell pose, lighting, and atmosphere"
}
}Template 6: Style transfer from a reference image
{
"reference": "use the uploaded image as the only source of identity, pose and framing",
"preserve": "face, facial proportions, hairline, garment silhouette, head angle",
"change": "rendering style only",
"target_style": "1970s screen-printed concert poster, three flat inks, visible halftone dots, slight misregistration between layers",
"palette": "burnt orange, deep teal, off-white paper",
"line_treatment": "hard posterised edges, no soft gradients, no airbrush shading",
"background": "replace with flat off-white paper texture and a thin printed border",
"text": "none",
"do_not": "do not restyle the face into a different person, do not change the head angle, do not add a signature",
"output": "same aspect ratio as the reference image"
}
The preserve and change pair is the whole template. Every style-transfer failure is the same failure — the model restyles the identity along with the surface — and the fix is to state which is which in two adjacent slots rather than hoping "keep the face" survives inside a paragraph.
Ultra Realistic Deconstruction Portrait
Nano Banana ProPrompt▾
3D Realistic Deconstruction Prompt. Based on reference image, create an ultra-high-quality, cinematic 3D realistic deconstruction art poster that emphasizes authentic human presence and photographic realism. Transform the photographed individual into a lifelike 3D character with true-to-life anatomy, skin texture, emotional nuance, and natural lighting. Arrange personal belongings using Knolling (orthogonal alignment) artistic layout. Aspect Ratio: 16:9 landscape. Artistic Style Core: Photorealistic Humanism — grounded in anatomical accuracy, natural subsurface scattering, visible pores, fine body hair, sweat sheen, and emotionally resonant realism. Quality Benchmark: Visually indistinguishable from high-end editorial photography. Total Items: 30–36 pieces arranged at precise 90-degree angles around the central figure. Categories include Fashion & Attire, Beauty & Self-Care, Intimate Private Life, Emotional Expression Gallery, Body Posture & Movement.
Three ways people break these templates
Leaving a slot with an empty value. Delete the key instead. A blank string is still text the model has to interpret, and it tends to be interpreted as an absence you wanted.
Writing two slots that contradict each other. "lighting": "hard direct flash" with "mood": "soft dreamy haze" does not produce tension, it produces an average. Contradictions belong in one slot, phrased deliberately: hard flash on a soft face.
Renaming slots and expecting a schema. There is no schema. lighting, light, illumination all work identically; the labels are for you, not the parser. What you cannot do is drop the slot and assume the concept is implied.
FAQ
Should I write AI image prompts in JSON?
Structured prompts show markedly higher median engagement than prose at the same length in our corpus — 141 vs 17 median views in the 80–149 word band. Strict JSON syntax is not the mechanism, though; labelled lines and numbered sections work the same way, because the benefit comes from named slots making an unfilled slot visible.
Do AI models actually parse JSON prompts?
They read it as text, not as data — there is no parser and no schema validation. The braces help you, not the model. That is why a malformed JSON prompt still works fine and why renaming a key changes nothing.
Can I put a negative prompt inside the template?
You can add a do_not or avoid key, as Template 6 does, and it is read as ordinary instruction text rather than a separate negative-prompt channel. Keep it to a short list of concrete things and put the positive statement in its own slot as well — "flat off-white background" works better than "no clutter" alone.
Which model handles structured prompts best?
We have no controlled test to answer that, so we will not claim one. What the corpus does show is a split in what people bring to each model: 51.7% of GPT Image prompts ask for text or typography, while 45.5% of Nano Banana prompts carry camera specifications. Template 3 suits the first, Templates 1 and 5 the second.
Where can I see more structured prompts?
The 1,249 structured prompts counted here are published with their images in the LocalBanana gallery, and any of them can be opened and run in the generator. All figures are counts over that public corpus as of August 2026.
LocalBanana Team
We run Nano Banana, GPT Image and other models side by side, and publish every prompt in our gallery. These guides are written from that corpus.
Updated @LocalBanana_io
