Tutorials8 min read
Keeping the Same Character Across AI Generations
Three layers that hold a character together across separate generations: an anchor description, one variable at a time, and reference images — plus when each stops working.
On this page
- The problem this solves
- Why characters drift
- Layer 1: Write the anchor once, paste it verbatim
- Layer 2: Change exactly one thing per generation
- Layer 3: Feed a reference image
- What the corpus says about this
- When it drifts anyway, fix in this order
- Character sets to start from
- FAQ
- How do I keep the same character in every AI image?
- Why does my AI character change between generations?
- Which model is best for consistent characters?
- Do I need a reference image, or is text enough?
- How is this different from a multi-panel character sheet?
The problem this solves
You made a character you like. The second image is a different person — same hair colour, wrong face. There is no seed to reuse and no character slot to save, so the only thing carrying identity from one generation to the next is the text you paste in.
This guide is about holding one character across separate generations: image one today, image nine next week. If what you need is several panels inside a single image — a six-expression sheet, a storyboard — that is a different mechanic and it has its own guide. Everything below assumes each render starts from scratch.
Why characters drift
A prompt is not a memory. Each generation re-derives a face from whatever description it was given, so any part of the face you did not specify gets re-invented — and "a young woman with long dark hair" leaves roughly everything unspecified.
Three things make drift worse, in order of impact:
- Vague anchors. Every adjective you leave out is a free variable. "Brown hair" spans six shades and four cuts.
- Paraphrasing between runs. Rewriting the anchor in your own words each time changes the render even when the meaning is identical. The text is your seed; treat it as bytes, not as prose.
- Changing more than one thing. New pose plus new outfit plus new lighting in one step, and you cannot tell which change moved the face.
The fix is three layers, applied in order. Most people only ever use the first one and conclude the model cannot do it.
Layer 1: Write the anchor once, paste it verbatim
The anchor is a fixed block of text describing the person, and nothing else. It never mentions the scene, the pose, the camera or the mood — those change every run, and anything that changes must live outside the anchor. That partition is the whole trick.
A usable anchor covers seven slots:
| Slot | Bad | Good |
|---|---|---|
| Age and build | young woman | 27, 168 cm, slim with broad shoulders |
| Face shape | pretty face | oval face, square jaw, high cheekbones |
| Hair | long brown hair | shoulder-length dark chestnut hair, blunt fringe, slight wave, parted left |
| Eyes | beautiful eyes | wide-set hazel eyes, heavy lower lids, thick straight brows |
| Skin | fair skin | warm olive skin, freckles across the nose, small mole under the right eye |
| Marker | — | one silver hoop in the left ear only |
| Wardrobe | casual clothes | cream ribbed knit, collar slightly stretched, thin gold chain |
The marker row is the one people skip and the one that does the most work. A single asymmetric, unusual detail — a septum ring, a scar through one eyebrow, mismatched earrings, a streak of dyed hair on one side — gives the model something it cannot average away, and gives you something you can check at a glance.
Elegant Anime Swordswoman 4-Panel Character Sheet
GPT ImagePrompt▾
A polished anime fantasy illustration sheet in a 2x2 grid showing the same elegant swordswoman in 4 panels. The character is a young woman with very long platinum-white hair tied in a high ponytail with large deep navy ribbon bows, soft loose curls and flowing strands, pale skin, and delicate refined features with luminous red-pink eyes. She wears an ornate white-and-navy gothic aristocratic dress: a white high-collar ruffled blouse with lace trim and gold embroidery, a large navy bow at the chest set with a red jewel brooch, puff sleeves with ribbon cuffs, a dark navy corset waist with gold details, and a voluminous navy skirt decorated with gold floral patterns, layered ruffles, and satin sheen. She carries a sheathed katana with a dark hilt and gold accents. Panel 1 is a close-up bust portrait from the chest up, turned slightly sideways, hair glowing in dramatic rim light against a dark sparkling night background. Panel 2 is a dynamic upper-body action shot, the character drawing or presenting the sword horizontally toward the viewer, hair streaming wildly, cinematic city-night bokeh and glowing petals around her. Panel 3 is a full-body fashion portrait, standing gracefully in a luminous reflective hall framed by tall glowing structures, showing the entire dress silhouette, heeled shoes with bows, and the sword at her side. Panel 4 is a three-quarter back view by a shimmering ocean at sunset or dawn, the character looking back over her shoulder, warm pastel sky and glittering water behind her, cherry blossom petals drifting in the breeze. Use ultra-detailed anime rendering, premium light novel cover quality, intricate fabric textures, glossy highlights, gold ornamentation, soft bloom, dramatic backlighting, floating petals, sparkling particles, elegant romantic atmosphere, and cohesive navy, white, gold, blush pink color harmony.
An anchor at full detail: hair colour, ribbon shape, eye colour, collar, lace trim, gold detailing — every one of those is a variable that would otherwise drift.Open in LocalBanana
Pixar Style Chibi Sticker Series
Nano Banana ProPrompt▾
A cute 3D chibi-style cartoon girl sticker set, featuring the same character in multiple emotions and poses. Big expressive glossy eyes, soft rounded face, slightly oversized head, small body proportions. Light brown/blonde messy bun hairstyle with loose strands. Wearing a red leather jacket over a white top. Each pose shows different expressions: surprised, sad, confused, happy, thinking, winking, peace sign, shrugging. Pixar/Disney-inspired 3D render, smooth plastic-like texture, soft lighting, pastel beige background. Sticker-style white outline around each character, clean and minimal aesthetic, high detail, ultra cute, emoji-like expressions, high resolution, studio quality.
One anchor plus a list of expressions. The character block is written once and the only thing that varies is the emotion.Open in LocalBanana
Here is the block to copy. Fill it once, save it in a text file, and paste the exact same characters into every future prompt.
[CHARACTER — do not edit between runs]
Name: Mara
Age and build: 27, 168 cm, slim with broad shoulders
Face: oval face, square jaw, high cheekbones, slightly asymmetric smile
Hair: shoulder-length dark chestnut, blunt fringe, slight wave, parted left
Eyes: wide-set hazel, heavy lower lids, thick straight brows
Skin: warm olive, freckles across the nose, small mole under the right eye
Marker: one silver hoop in the left ear only, none in the right
Wardrobe: cream ribbed knit, collar slightly stretched, thin gold chain
[SCENE — the only part that changes]
...
Layer 2: Change exactly one thing per generation
With the anchor fixed, the second layer is discipline about what varies. One variable per run: the pose, or the location, or the outfit, or the light — never two.
✕ Bad
same girl as before, now in a cafe, wearing a red dress, sunset light, laughing, from a low angle✓ Good
[paste the identical CHARACTER block] SCENE: seated at a cafe window table, mid-morning, same cream knit as before, calm expression, eye level, 50mm. Everything about the character is unchanged from the previous image.Two extra habits are worth copying. First, state what must not change, explicitly, at the end — redundant to a human reader, load-bearing for a model. Second, if the character matters more than the scene, say so with a priority word: the character sets in our library that span a whole grid tend to declare identity as the top constraint before anything else, rather than burying it mid-paragraph.
9-Grid Editorial × Bare 3D Pop-Out Fashion Composition
Nano Banana ProPrompt▾
Create a 2:3 portrait fashion poster featuring THE SAME WOMAN in THE SAME OUTFIT shown in 9 different magazine editorial styles with 3D pop-out effect: CHARACTER CONSISTENCY (CRITICAL - HIGHEST PRIORITY): THE SAME female fashion model appears in ALL 9 positions: - Same face, same facial features, same skin tone, same body type - Cold-beauty aesthetic: sharp jawline, high cheekbones, aloof minimalist expression - Early-20s Chinese/Korean fashion model with editorial face - Her identity NEVER changes across all 9 appearances OUTFIT CONSISTENCY (NEW RULE): THE SAME OUTFIT in all 9 positions: - Oversized black cashmere V-neck sweater (slightly loose fit) - High-waisted wide-leg pure white tailored trousers - Black leather loafers with subtle gold horsebit detail - Neat low bun with slightly messy front strands - Small gold hoop earrings, thin gold chain necklace SAME CLOTHING - only photography style, pose, and angle vary BACKGROUND LAYER (Z=0) - 3×3 Grid with 8 Visible Magazine Styles: Grid Structure & Occlusion: - Standard 3×3 layout = 9 magazine editorial shots - **8 visible cells** (center cell [2,2] COMPLETELY OCCLUDED by 3D figure) - Cells separated by DISTINCT THICK WHITE LINES (3-4px) for clear separation [1,1] Vogue Editorial Style: - Same woman, same outfit - Pose: Standing tall, hand in pocket, direct powerful gaze - Style: High contrast lighting, dramatic shadows, sophisticated - Sharp focus, clear face [1,2] Harper's Bazaar Style: - Same woman, same outfit - Pose: Side profile, looking over shoulder - Style: Soft glamour lighting, elegant mood - Sharp focus, clear face [1,3] Elle Street Style: - Same woman, same outfit - Pose: Walking motion, casual confident stride - Style: Natural daylight, urban chic aesthetic - Sharp focus, clear face [2,1] i-D Magazine Style: - Same woman, same outfit - Pose: Sitting on minimal cube, legs crossed - Style: Bold graphic composition, colorful backdrop - Sharp focus, clear face [2,3] Dazed & Confused Style: - Same woman, same outfit - Pose: Dynamic movement, fabric flowing - Style: Experimental angles, artistic editorial - Sharp focus, clear face [3,1] Marie Claire Corporate Chic: - Same woman, same outfit - Pose: Power stance, arms crossed professionally - Style: Clean corporate aesthetic, neutral tones - Sharp focus, clear face [3,2] GQ Minimalist Style: - Same woman, same outfit - Pose: Leaning against wall, relaxed elegance - Style: Architectural composition, clean lines - Sharp focus, clear face [3,3] W Magazine Avant-Garde: - Same woman, same outfit - Pose: Artistic pose, hand gestures expressive - Style: Bold contrast, fashion-forward editorial - Sharp focus, clear face CRITICAL TECHNICAL SPECS FOR BACKGROUND GRID: - Deep depth of field (f/16) - ALL faces sharp and clear - NO bokeh, NO blur, NO out-of-focus areas - Even bright studio lighting across all cells - High resolution faces in every cell - Thick white grid lines clearly visible between cells - Background color: Bright minimalist concrete/white studio FOREGROUND LAYER (Z=5-10cm forward) - Hyper-Realistic 3D Pop-out: THE SAME WOMAN, SAME OUTFIT (Look 5 - Most Dramatic): - Massive hyper-realistic full-body shot dominating the center - Positioned at EXACT CENTER, completely occluding center cell [2,2] - **Head touches very top edge of canvas** - **Shoes touch very bottom edge of canvas** - Occupies MAXIMUM vertical space for strong 3D illusion Pose: - Dynamic walking forward motion - Confident stride, mid-step - Hand on hip or naturally swinging - Direct gaze at camera, commanding presence - Full body visible from head to toe Technical Execution: - Figure extends 5-10cm forward from background plane - Hyper-realistic detail (skin texture, fabric weave visible) - +20% saturation compared to background for "pop forward" effect - Slightly sharper focus than background (but background still sharp) OCCLUSION MECHANICS (9格 - 1格遮挡 = 8格可见): Complete Occlusion: - Figure's body COMPLETELY covers center cell [2,2] (100% invisible) - Center magazine shot is fully hidden behind 3D figure Partial Occlusion (Natural Edge Overlap): - Top [1,2]: Hair/head overlaps 10-15% into Harper's Bazaar shot - Left [2,1]: Left arm/sleeve overlaps 15-20% into i-D shot - Right [2,3]: Right arm overlaps 15-20% into Dazed shot - Bottom [3,2]: Legs/feet overlap 10-15% into GQ shot - Overlaps break the white grid boundaries naturally Edge Treatment: - Soft organic transitions, NO hard cutout edges - Figure appears to physically exist in front of the grid - Like a 3D cardboard cutout standing in front of a poster DEPTH EFFECTS: Shadows: - Drop shadow from 3D figure onto grid background * Blur: 12px * Color: rgba(0,0,0,0.25) (slightly darker for stronger effect) * Offset: X=6px, Y=10px - Contact shadow where figure "stands" on background * Blur: 8px * Color: rgba(0,0,0,0.35) * Creates grounding effect Lighting: - Background grid: Even bright studio lighting (no dramatic shadows) - Foreground figure: * Key light upper left 45° * Subtle rim light on edges for separation * Slightly more dramatic lighting than background - Consistent lighting direction across all elements Separation Techniques: - Slight brightness difference (foreground +10% brighter) - Slight saturation boost (foreground +20% more saturated) - Subtle sharpening halo around figure edges - Clear Z-axis spatial hierarchy CONSISTENCY RULES (ABSOLUTE PRIORITY): Same Woman Verification: - Same face in all 9 positions - Same facial structure, eyes, nose, lips, jawline - Same cold-beauty editorial expression - Same hair styling (low bun, messy strands) - Same age, same ethnicity, same beauty Same Outfit Verification: - Same black sweater in all 9 shots - Same white trousers in all 9 shots - Same accessories (earrings, necklace, loafers) - Only photography style and pose differ What Changes: - ✅ Magazine editorial style (lighting, mood, composition) - ✅ Pose and body angle - ✅ Camera angle and framing - ✅ Photographic treatment What NEVER Changes: - ❌ The woman's face or identity - ❌ The outfit or clothing items - ❌ The accessories - ❌ The overall styling concept TECHNICAL SPECIFICATIONS: Image Composition: - Aspect ratio: 2:3 portrait (or 9:16 vertical) - Resolution: 2000×3000 pixels (or higher) - Color mode: RGB, sRGB color space - Quality: Professional editorial fashion photography Camera & Focus: - **Deep depth of field (f/16 or higher)** - **NO selective focus, NO bokeh, NO blur** - **ALL faces in background grid MUST be sharp and clear** - Foreground figure slightly sharper for hierarchy - Both layers fully illuminated and visible Environment: - Bright minimalist indoor studio - Concrete walls or pure white background - Optional: Minimal green plants for visual interest - Clean, uncluttered aesthetic - Quiet luxury mood Layout: - Background: Clear 3×3 grid with THICK WHITE LINES visible - Foreground: Massive full-body figure breaking grid boundaries - Surreal creative collage composition - Graphic and editorial feel FORBIDDEN ELEMENTS (严格禁止): Character & Outfit: - ❌ Different women in different cells - ❌ Different outfits or clothing changes - ❌ Changing facial features or styling - ❌ Multiple models instead of one person Technical: - ❌ Blurred background or bokeh effect - ❌ Out of focus faces in grid - ❌ Shallow depth of field - ❌ Missing or unclear grid lines - ❌ Dark shadows obscuring faces - ❌ Low resolution or pixelation - ❌ Deformed limbs or merging bodies - ❌ Messy composition Structure: - ❌ 4×4 or other grid sizes (must be 3×3) - ❌ All 9 cells visible (center must be occluded) - ❌ Flat composition (must have clear 3D depth) - ❌ Hard cutout edges on foreground figure QUALITY CHECKLIST: Before Generation: - [ ] Same woman's face in all 9 positions? - [ ] Same outfit in all 9 positions? - [ ] Each cell shows different magazine editorial style? - [ ] Center cell [2,2] completely hidden? - [ ] 8 visible background cells clearly defined? - [ ] Thick white grid lines visible? - [ ] ALL background faces sharp and clear (no blur)? - [ ] Foreground figure full-body, head-to-toe? - [ ] Figure extends maximum vertical space? - [ ] Clear 3D pop-out effect? - [ ] Natural edge overlaps into adjacent cells? - [ ] Shadows present for depth? - [ ] Deep depth of field maintained? MIDJOURNEY/AI COMMAND FORMAT: /imagine prompt: A surreal 3x3 fashion grid collage with THICK WHITE LINES separating cells. Background shows THE SAME Chinese fashion model in THE SAME black oversized sweater and white wide-leg trousers in 8 different magazine editorial styles (Vogue, Harper's Bazaar, Elle, i-D, Dazed, Marie Claire, GQ, W Magazine) - various poses but identical outfit. CENTER CELL HIDDEN. OVERLAID by a massive hyper-realistic full-body 3D cut-out of the SAME MODEL in SAME OUTFIT walking forward, head touching top edge, feet touching bottom edge. ALL faces sharp and in focus, deep depth of field f/16, no blur anywhere, bright studio lighting, clear white grid lines, strong 3D pop-out effect, editorial photography, same woman same clothes 9 times, 8k resolution --ar 2:3 --v 6.1 --stylize 300 --quality 2 MATHEMATICAL LOGIC: Same woman × Same outfit × 9 different magazine editorial styles arranged in 3×3 grid. Center style completely occluded by 3D foreground version = 8 visible background editorial styles + 1 foreground 3D editorial = 9 total appearances of ONE PERSON in ONE OUTFIT with NINE photographic interpretations.
Character consistency declared as the highest-priority constraint before anything else, with the outfit pinned alongside the face.Open in LocalBanana
Stickers of Different Clothing Styles
Nano Banana ProPrompt▾
A 3x3 sticker pack based on an uploaded reference photo, featuring the same character wearing a variety of different outfits and fashion styles. Cleanly cut edges, bold thick-line outlines, and expressive, dynamic poses, with an overall lively and modern sticker design style. Each sticker includes bilingual labels in both Chinese and English placed next to the character. All stickers maintain identical facial features, consistent likeness, proportions, and scale across the entire set. The sticker pack includes multiple outfit styles such as: teacher attire, traditional clothing, nurse uniform, streetwear fashion, and fantasy-inspired outfits. High-resolution final output, with soft drop shadows and a glossy sticker paper texture, suitable for social media sharing.
Outfit is the variable, identity is the invariant — the prompt states that facial features, likeness and proportions stay identical across every sticker.Open in LocalBanana
Layer 3: Feed a reference image
Text can pin a character to about the level of "recognisably the same person". A reference image pins it to the same face. If you have one usable render of the character, every subsequent generation should start from it rather than from the description alone.
The instruction matters as much as the upload. These three patterns appear throughout our library and each does a different job:
- Strict identity lock — "Use reference image. Preserve identity and facial structure strictly." Use when the face must survive a change of scene, wardrobe or lighting.
- No-stylisation clause — "Preserve facial proportions, skin texture, expression and identity features with 100% accuracy. Do not stylise or alter facial features." Use when the rest of the prompt asks for a strong art style that would otherwise pull the face with it.
- Placeholder slot — writing the subject as a named reference token that the upload fills in. Useful when you reuse one prompt template across different people.
Controlled Gaze Under Dominant Flash
Nano Banana ProPrompt▾
Use reference image. Preserve identity and facial structure strictly. A slim Asian woman standing still, body deliberately composed rather than casual. Her posture feels arranged and intentional, not relaxed. She faces the camera directly, shoulders stable, center of gravity controlled. Her expression is calm and restrained — not playful, not overly sweet. A controlled, subtle smile, with emotional distance. Her gaze is steady and aware, fixed toward the camera, creating quiet dominance rather than friendliness. Hands & Props Right hand holding a hanging cluster of golden trumpet flowers, positioned deliberately as a visual accent Left hand holding a larger bouquet of golden trumpet flowers, lowered slightly Flowers function as props only, never overpowering the subject The woman remains the absolute visual focal point at all times Hair Long hair, slightly messy layered style Hair gently lifted by wind, with a few loose strands crossing the face Chic, slightly sexy, but controlled — never messy or cute Outfit White vintage-style puff-sleeve blouse Large puffed sleeves Layered ruffle cuffs Square neckline Drawstrings at the chest forming soft gathers Slightly long length covering the hips Front slit visible but restrained Light blue jeans Colorful beaded necklace as a subtle accent Makeup Fair skin tone Douyin + Korean beauty inspired Long curled eyelashes Soft pink blush on cheeks and nose tip Glossy pink-orange lips with juicy shine Makeup feels polished and deliberate, not cute Nails White French tip nails Subtle Christmas-themed nail art, low contrast Environment Standing directly in front of a fully blooming Golden Trumpet Tree (Thong Urai) The tree is dense with bright golden flowers, forming a saturated backdrop Some flower stems partially overlap the top edge of the frame, creating a soft foreground blur Background exists to frame the subject, not to compete with her Lighting (CRITICAL) Direct flash is the dominant key light Flash visibly overpowers ambient light Skin appears very bright with slight yellow flash tone Clear, hard-edged shadow cast behind the subject Natural sunlight exists only as weak ambient fill Lighting creates strong separation between subject and background Photography Style Compact camera aesthetic Fujifilm Pro 400H film look Vintage lens character Subtle film grain Ultra-sharp subject clarity Bright exposure with aggressive subject emphasis 8K resolution The image feels intentional, staged, and directed — not casual or spontaneous
Two sentences of identity lock at the very top of the prompt, before a single word about pose or light.Open in LocalBanana
Elves in the Snowscape
Nano Banana ProPrompt▾
Editorial winter poster style multi-panel collage with spontaneous iPhone photography aesthetic (candid, warm, realistic). Soft snowflakes with delicate analog grain and slight handheld imperfections. Reference Adherence: Strictly follow the provided photo for facial reference with zero deviation. Preserve facial proportions, skin texture, expression, and identity features with 100% accuracy. Do not stylize or alter facial features. Consistent Elements: - Subject Wardrobe: Short faux fur coat, black leggings, classic UGG boots (minimalist, cozy, very wintry style) - Primary Device: iPhone 17 Pro Max in silver, held by the subject in relevant frames - Color Palette: Warm amber, soft red, pine green, soft winter gray Layout Configuration: Panel 1 (Top Left): Store window reflection photo at dusk. Faint Christmas lights, garlands, frosted glass edges, warm highlights on fur. Subject holding phone partially covering face. Grazing silhouettes, layered reflections, soft ghosting, natural glass curvature distortion. Panel 2 (Top Right): Ultra-wide street scene portrait (snowy sidewalk/Christmas market). Close-up, tilted downward camera angle. Subject casually leaning forward, hands in coat pockets. Black leggings and UGG boots clearly visible. Falling snow with slight motion blur. Subtle perspective distortion to enhance handheld authenticity. Panel 3 (Bottom Right): Intimate overhead selfie with warm street or cafe lighting. Holding a takeaway festive drink (coffee or mulled wine). Visible wired earphones. Clear detailed fur texture and winter fabrics. Soft grain enhancing nostalgic holiday atmosphere. Graphic Overlay: Minimalist Apple Music style mini player floating in the center of the collage, showing a popular Christmas song (e.g., 'Last Christmas' or 'All I Want for Christmas Is You'). Flat clean render without shadows.
A heavy stylistic treatment with an explicit do-not-stylise clause on the face — the style is allowed to touch everything except identity.Open in LocalBanana
One Image Generates 9 Different Shots
Nano Banana ProPrompt▾
<instruction> Analyze the entire composition of the input image. Identify all key subjects present (whether a single person, group/couple, vehicle, or specific object) and their spatial relationships/interactions. Generate a coherent 3x3 grid "Contact Sheet" showcasing 9 distinct shots of exactly these subjects within the same environment. You must adapt standard cinematic shot types to fit the content (e.g., if a group, keep the group together; if an object, frame the entire object): Row 1 (Establishing Context): 1. Extreme Long Shot (ELS): Subjects appear very small within a vast environment. 2. Long Shot (LS): Full subject or group visible from top to bottom (head to toe / wheels to roof). 3. Medium Long Shot (American Shot/Three-Quarter): Framed from above the knees (for people) or a 3/4 view (for objects). Row 2 (Core Coverage): 4. Medium Shot (MS): Framed from above the waist (or the central core of an object). Focus on interaction/action. 5. Medium Close-Up (MCU): Framed from above the chest. Intimate framing of the primary subject(s). 6. Close-Up (CU): Tight framing on the face or the "front" of an object. Row 3 (Detail & Angles): 7. Extreme Close-Up (ECU): Macro detail with intense focus on a key feature (eyes, hands, logo, texture). 8. Low-Angle Shot (Worm's-Eye View): Looking up at the subject from ground level (spectacular/heroic feel). 9. High-Angle Shot (Bird's-Eye View): Looking down on the subject from above. Ensure strict consistency: The same people/objects, same clothing, and same lighting across all 9 panels. Depth of field should vary realistically (background blur in close-ups). </instruction> A professional 3x3 cinematic storyboard grid containing 9 panels. This grid showcases the specific subject(s)/scene from the input image across a comprehensive range of focal lengths. Top Row: Wide environmental shots, full view, 3/4 crop (above-knee view). Middle Row: Above-waist view, above-chest view, face/front close-up. Bottom Row: Macro detail, low angle, high angle. All frames feature photorealistic textures, consistent cinematic color grading, and correct framing for the specific number of subjects or objects analyzed.
Nine new shots derived from one input image: the reference supplies the subject, the prompt supplies only the camera work.Open in LocalBanana
Indie Sleaze Cool Girl Flash Portrait
Nano Banana ProPrompt▾
{
"portrait_prompt": {
"subject": "Based on <User Portrait>, a young woman with voluminous, messy, long hair, wearing a black sporty crop top with white trim and matching black shorts, white calf-high socks. She is sitting perched on a high dark wooden counter or piano top. Her pose is casual and edgy, leaning forward with one hand resting near her mouth, biting her finger slightly, gazing directly at the camera with a sultry, nonchalant expression. A silver bracelet on her wrist. Background includes a white wall with vintage posters taped up and brown wooden blinds. Foreground details include the neck of an electric guitar, a glass jar of cookies, and liquor bottles",
"composition": "Full body shot, 35mm focal length, slightly low angle to emphasize leg length, sharp focus on subject with hard flash fall-off, messy but balanced framing",
"camera_angle": "Eye-level relative to the seated subject, slightly low angle from the floor, medium distance",
"lighting": "Direct on-camera flash photography, hard lighting creating a sharp drop shadow on the wall behind the subject, high contrast, reminiscent of 90s point-and-shoot aesthetics, no diffusion",
"color_palette": "Vintage film aesthetic, Kodak Gold 200 simulation, warm tones from wooden blinds and furniture contrasted with cool white flash light, deep blacks, slightly grainy texture, lo-fi indie vibe",
"mood": "Indie sleaze, candid, rebellious, cool girl aesthetic, raw and authentic, Y2K retro fashion editorial, snapshot style",
"negative_prompt": "soft lighting, studio lighting, bokeh, professional studio portrait, airbrushed skin, 3d render, cartoon, illustration, distorted hands, missing guitar strings, floating objects, anatomical errors, stiff pose, over-processed, HDR"
}
}The placeholder pattern — the subject slot names a supplied portrait, so the same prompt can be reused for a different person unchanged.Open in LocalBanana
What the corpus says about this
Counted across our own gallery: 9,667 published prompts, 9,599 of them longer than 20 characters, as of August 2026. These are usage frequencies, not a benchmark.
11.9%
of Nano Banana prompts work from a reference or uploaded image
GPT Image 5.1%, Midjourney 0.2%
4.8%
of Nano Banana prompts contain consistency language
GPT Image 4.2%, Midjourney 0.3%
9.4%
of all prompts request a grid, panel set or character sheet
901 of 9,599 prompts
The gap in the first row is the practical story. Reference-driven prompting is routine on Nano Banana and effectively absent on Midjourney — 7 prompts out of 4,249 — and the consistency wording follows the same split. That is a workflow people adopted where it worked and abandoned where it did not. Conditional instructions of the "keep this, change only that" kind are where Nano Banana separates in our corpus, which is why nearly every character-consistent set in the gallery was made with it. The rest of the model split is in our 9,667-prompt breakdown.
Nano Banana Pro
Pixar-Style 6-Panel Expression Grid
GPT Image
Cat-Eared Student Council Character Reference
When it drifts anyway, fix in this order
- Check the marker first. If the asymmetric detail survived and the face still looks off, the problem is your face description, not the model.
- Re-paste the anchor from the file. Nine times out of ten the anchor got paraphrased somewhere along the way.
- Cut back to one variable. Return to the last render that worked and change one thing.
- Move identity to the top and mark it critical. Position in the prompt matters; buried constraints get outvoted.
- Add the reference image. Strongest lever, and the reason to keep your best render of the character somewhere findable.
- Reduce the ask. Nine poses in one image drifts more than three images of three poses — resolution per face is a real constraint.
Nine-Grid Hippocampus Studio Portrait
Nano Banana ProPrompt▾
masterpiece, best quality, photorealistic, ultra-high resolution, vibrant colors, professional MV studio photography, a 3x3 photo grid collage. A collection of 9 glamorous portraits of the same person: a stunningly beautiful K-pop idol in her early 20s. She has large, captivating double-lidded eyes, flawless porcelain skin, and long, dark, voluminous wavy hair. She is wearing an elegant, vibrant red strapless dress. Her makeup is immaculate, inspired by a K-pop music video: shimmering eyeshadow, sharp eyeliner, and glossy, gradient lips. Each grid cell captures a different charismatic expression and pose: 1. **Top-left:** Taking a selfie, holding a phone just out of frame, with a sweet, fan-service smile. 2. **Top-center:** A playful wink with one eye closed and a charming smile. 3. **Top-right:** A surprised expression with wide eyes and an open mouth in an "O" shape, looking adorable. 4. **Middle-left:** A cute pouting face with pursed lips, head tilted slightly, showing her "aegyo" (cuteness). 5. **Middle-center:** A joyful, radiant smile with eyes crinkled shut, exuding pure happiness. 6. **Middle-right:** A mischievous, scrunched-up face, wrinkling her nose playfully. 7. **Bottom-left:** Looking back over her shoulder, playfully sticking the tip of her tongue out. 8. **Bottom-center:** A big, happy, open-mouthed laugh, looking joyfully upwards like a candid moment. 9. **Bottom-right:** A shy, gentle smile while looking away to the side, creating a "first love" feeling. Flawless, bright, and dynamic studio lighting, mimicking a high-budget music video set. Uses a combination of softboxes for a clean base and a subtle key light to add dimension and a glamorous sheen to her skin and hair. Clean, seamless, solid light gray studio background to make the red dress pop. Captured with a high-end portrait lens (equivalent to an 85mm f/1.2), creating a beautiful, creamy bokeh that makes the subject stand out sharply. The palette is rich and saturated, with the vibrant red of the dress as the focal point. The overall mood is charismatic, polished, and full of star quality.
Nine portraits of one person in a single frame. At this density the anchor has to carry hair, eyes, dress and makeup, or panel seven becomes someone else.Open in LocalBanana
Realistic faces drift more than stylised ones
A cel-shaded character with a distinctive silhouette survives almost anything. A photorealistic face has thousands of tiny degrees of freedom and drifts on all of them. If you are working photographic, expect to need the reference image, not just the text — and see the realism guide for how to keep skin from being smoothed into a stranger.
Character sets to start from
Open any of these, copy the prompt, and try swapping only the character block — that is the fastest way to see which parts of a prompt are anchor and which are scene. More in the prompt library, and you can run any of them directly in the generator.
FAQ
How do I keep the same character in every AI image?
Write one fixed character block — age, build, face shape, hair, eyes, skin, one unusual asymmetric marker, wardrobe — and paste the identical text into every generation. Change one thing per run, state at the end that the character is unchanged, and supply a reference image once you have a render you like.
Why does my AI character change between generations?
Because nothing carries over. Each render rebuilds the face from the description, so every detail you did not write becomes a free variable. Paraphrasing the description between runs has the same effect as changing it.
Which model is best for consistent characters?
In our library, Nano Banana: 11.9% of its prompts work from a reference image against 0.2% for Midjourney, and conditional "keep this, change that" instructions are its clearest strength. GPT Image is the better pick when the character sheet itself needs readable labels or text, which you can browse in the GPT Image prompt hub.
Do I need a reference image, or is text enough?
Text alone gets you to "recognisably the same person" — good enough for stylised or illustrated characters. For photorealistic faces, plan on the reference image. It is the single strongest lock available and it frees the prompt to describe scene and light instead of bone structure. It is also what makes a style change survivable: ask for the new style everywhere, and explicitly exempt facial proportions and identity from it.
How is this different from a multi-panel character sheet?
A sheet solves consistency inside one image, where the model can see all panels at once — that is the easier problem, covered in the multi-panel guide. This page is about consistency across separate generations, where the model sees nothing but the text you paste in.
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
