Introduction
Cette “influenceuse” n’est pas un compte réel à décrire : c’est une identité visuelle à recréer (et décliner) à partir d’un portrait de référence.
L’objectif : obtenir une personne reconnaissable (même visage, mêmes proportions, même “vibe”) dans des scènes très différentes
(street look, musée, night-out, dîner, plage), sans dérive (“drift”).
Résultats (exemple — sortie du process)
Voici un rendu généré avec ce workflow (portrait de référence + prompt structuré). C’est un exemple de “résultat” à afficher dans l’article.
Résultat — sortie du process (REF portrait + prompt “influencer_generation”)
ADN de l’influenceuse (ce qu’on “lock”)
- Identité : visage + proportions (c’est le rôle de la photo de référence).
- Signature look : coiffure sleek low bun + raie au centre (dans 5/6 images), bijoux or discrets, makeup éditorial naturel.
- Rendu : “photo réelle” (grain iPhone / imperfections / composition non publicitaire).
- Règle d’or : on change une seule grande chose à la fois (lieu OU tenue OU cadrage), sinon drift.
Le prompt (structure “template” à réutiliser)
Ce format marche très bien avec un modèle image/édition type “Nano Banana Pro” (Gemini) et se décline ensuite en photoshoot.
Le point clé : subject_reference = la personne fournie en entrée (portrait).
{
"task": "influencer_generation",
"subject_reference": "Use the person provided in the input portrait",
"scene_description": "A sophisticated young woman with hair styled in a sleek low bun with a polished finish and a clean centre par. The moment feels personal and unplanned, as if the photo is taken by her friend sitting across the table, she is seated casually at a table inside a dimly lit upscale restaurant during dinner, she rests one arm on the table while the other hand is loosely lifted near her face looking directly toward the camera with a soft, attentive expression, as if responding to someone seated across the table, creating an intimate date-night moment. The table setting feels naturally imperfect and lived-in, there is a standard fine-dining water glass with thin walls, a symmetrical cylindrical shape, and a flat even base placed slightly off-center, there is a red wine glass that is visibly half-finished with the wine level, a folded or partially opened menu resting casually nearby, there is a small tea light candle. The overall feeling is joyful, unguarded, and real capturing a sequence of fleeting date-night moments.",
"outfit": "Elegant black strapless evening dress with a structured bodice and clean lines, draping naturally over the body. Styled simply and without emphasis.",
"style": {
"vibe": "Intimate, high-end casual snapshot, accidental renaissance aesthetic, soft glamour, moody aesthetic",
"color_palette": "Warm browns, muted golds, soft beige, black, natural skin tones",
"look": "Soft focus, cinematic grain (digital noise), non-commercial, authentic lifestyle image"
},
"composition": {
"shot_type": "Medium close-up, slightly low angle from table level",
"framing": "Tight but natural crop with breathing room; subject framed with visible table context and surrounding space",
"pose": "Relaxed seated posture with forearms resting on the table; shoulders soft; head gently oriented toward the camera with calm eye contact",
"camera_constraint": "Camera height and angle must remain fixed at seated table level, slightly below the subject’s eye line; no automatic elevation to eye-level"
},
"technical": {
"lighting": "Mixed low-key lighting, dominated by warm incandescent lamps and candlelight, creating strong rim lighting and soft skin highlights. Avoid HDR.",
"camera": "Shot on iPhone 15, vertical orientation (4:5 or 3:4 aspect ratio), natural lens flare from light sources, high ISO noise visible in shadows.",
"makeup": "Clean, natural editorial makeup with softly defined brows, minimal eye makeup, natural lashes, gentle warmth in the cheeks, subtle skin luminosity, and a satin-finish lip.",
"hair": "Styled in a sleek low bun with a polished finish and a clean center part, remaining neat throughout the moment."
},
"environment": {
"location": "Upscale, traditional dining room with a private club aesthetic",
"details": "Deep burgundy velvet drapery, dark polished mahogany wood paneling, brass chandelier fixtures, floral upholstered chairs, white linen tablecloths, antique pleated table lamps with warm yellow light.",
"table_variation_rules": {
"placement_variability": "Table objects are not placed in fixed or repeatable positions; their placement varies subtly between moments as in real dining.",
"non_symmetry": "Avoid symmetrical or templated arrangements; placement should feel incidental, not designed.",
"constraint": "All variation must remain physically plausible and within the same table plane."
},
"jewelry_rendering": "Jewelry appears softly reflective under ambient light, present but understated and slightly imperfect"
},
"physics_override": {
"gravity_reference": "World gravity is vertical regardless of subject pose or camera tilt",
"liquid_rule": "All liquids remain level relative to gravity, not camera or subject",
"object_geometry": "Glassware and tables obey real-world optics and geometry even if the camera is slightly tilted"
}
}
Décorticage (comment cette structure te permet de refaire toutes les images)
- task : juste un label pour ton workflow (utile si tu automatises dans n8n/DB).
- subject_reference : le verrou principal. Tu fournis un portrait input, le modèle doit conserver cette identité.
- scene_description : le “film” de la scène (actions, micro-détails, ambiance). C’est le bloc à modifier quand tu changes de lieu.
- outfit : décrit la tenue en termes simples + contraintes (ex: “without emphasis” = évite le rendu pub/sexualisé).
- style : ton LUT verbal :
- vibe = intention (snapshot, luxe, moody, etc.)
- color_palette = cohérence de grading entre images
- look = grain/imperfections (très important contre l’effet “IA trop propre”)
- composition : évite les sorties “automatiques” :
- shot_type + framing = cadrage stable
- pose = langage corporel
- camera_constraint = la phrase anti-drift n°1 (sinon le modèle remonte la caméra, change la focale, etc.)
- technical : rend la photo crédible (et répétable) :
- lighting : low-key / HDR off = contraste réaliste
- camera : “Shot on iPhone … vertical … high ISO noise” = texture
- makeup/hair : stabilise les détails identitaires (sourcils, glow peau, chignon)
- environment : décor + règles de variation “réaliste” (objets pas symétriques, pas de mise en scène trop parfaite).
- physics_override : petit bloc, énorme impact : verres/liquides/angles restent physiquement plausibles (sinon “IA vibes”).
Workflow recommandé (pour refaire ces 6 images avec n’importe quelle personne en référence)
- Étape 1 — Choisis 1 portrait REF : visage net, lumière neutre, pas d’accessoires qui masquent (lunettes) si possible.
- Étape 2 — Écris un “Identity Pack” (invariant) : hair/makeup/jewelry signature + “non-commercial snapshot”. Tu le réutilises partout.
- Étape 3 — Déclines par “Scene Pack” : tu changes seulement scene_description + environment + outfit + composition.
- Étape 4 — Itère en micro-variations (3 à 6 rendus) : mêmes contraintes caméra, petits changements (regard, main, distance, placement objets).
- Étape 5 — Si drift : re-baisse l’ambition (moins d’objets, moins d’actions), ou ajoute une 2e référence (style/tenue/lieu) en plus de l’identité.
Prompts “Scene Pack” (6 scènes prêtes à copier-coller)
Même structure, mêmes invariants (identity/hair/makeup) — tu remplaces uniquement les blocs ci-dessous.
Scene Pack 01 — Day look / devant une porte
{
"task": "influencer_generation",
"subject_reference": "Use the person provided in the input portrait",
"scene_description": "A casual daytime street-style photo in front of a tall dark wooden door framed by light stucco walls and a shaded entryway. The subject stands relaxed, facing camera, natural posture, calm expression. Terracotta tile floor, soft midday ambient light, realistic shadows, non-commercial candid feel.",
"outfit": "Green camisole top with subtle lace trim, burgundy ruffled mini skirt with a small floral print, dark brown knee-high leather boots, small black shoulder bag.",
"style": {
"vibe": "Authentic lifestyle, casual chic, natural",
"color_palette": "Warm neutrals, deep green, burgundy, brown leather",
"look": "Slight softness, realistic texture, no studio feel"
},
"composition": {
"shot_type": "Full body portrait",
"framing": "Subject centered, head-to-boots visible, straight vertical lines, 3:4",
"pose": "Relaxed arms, slight weight shift, direct gaze",
"camera_constraint": "Eye-level to slightly below, fixed focal length, no extreme wide-angle distortion"
},
"technical": {
"lighting": "Soft natural daylight in shade, no HDR, realistic contrast",
"camera": "Shot on iPhone, vertical 3:4, natural sharpness, mild noise",
"makeup": "Clean editorial natural makeup, softly defined brows, subtle glow",
"hair": "Sleek low bun with clean center part"
},
"environment": {
"location": "Residential entryway",
"details": "Large wooden door, light stucco walls, terracotta floor tiles"
},
"physics_override": {
"gravity_reference": "World gravity is vertical regardless of camera tilt"
}
}
Scene Pack 02 — Night selfie / ville + flash
{
"task": "influencer_generation",
"subject_reference": "Use the person provided in the input portrait",
"scene_description": "A handheld night selfie on a city street at blue hour. Background is softly blurred with streetlights and buildings, subtle motion blur, candid nightlife vibe. The subject looks into the camera with a soft confident expression.",
"outfit": "Minimal black spaghetti-strap dress, small gold hoop earrings, delicate gold chain necklace.",
"style": {
"vibe": "Night-out snapshot, moody, real iPhone flash",
"color_palette": "Deep blues, warm streetlight bokeh, natural skin tones",
"look": "Direct flash highlights, visible skin texture, non-commercial"
},
"composition": {
"shot_type": "Tight selfie close-up",
"framing": "Head and shoulders, arm slightly visible, 3:4",
"pose": "Neutral head angle, relaxed mouth, calm gaze",
"camera_constraint": "Handheld selfie perspective, slight tilt allowed, no perfect studio symmetry"
},
"technical": {
"lighting": "On-camera flash + ambient city light, avoid HDR, mild noise in shadows",
"camera": "Shot on iPhone 15, vertical 3:4, shallow-ish depth from phone processing",
"makeup": "Natural editorial makeup, satin lip",
"hair": "Sleek low bun with clean center part"
},
"environment": {
"location": "City street at night",
"details": "Crosswalk lines, distant buildings, bokeh lights"
},
"physics_override": {
"gravity_reference": "World gravity is vertical regardless of camera tilt"
}
}
Scene Pack 03 — Galerie d’art / blazer
{
"task": "influencer_generation",
"subject_reference": "Use the person provided in the input portrait",
"scene_description": "A modern minimalist art gallery with white walls and soft ceiling spotlights. The subject takes a casual selfie, slightly turned in profile, calm thoughtful expression. An abstract black-and-white artwork is visible in the background.",
"outfit": "Taupe/beige tailored blazer over a simple black top, black shoulder bag, understated gold jewelry.",
"style": {
"vibe": "Quiet luxury, museum day, editorial but candid",
"color_palette": "Soft whites, beige/taupe, black accents",
"look": "Clean sharp image, subtle softness, realistic lighting"
},
"composition": {
"shot_type": "Medium close-up selfie",
"framing": "Upper torso and head, gallery context visible, 3:4",
"pose": "Shoulders relaxed, gaze slightly forward, not exaggerated",
"camera_constraint": "Phone held slightly forward, no cinematic camera moves"
},
"technical": {
"lighting": "Even diffused gallery light, no HDR punch",
"camera": "Shot on iPhone, vertical 3:4, realistic sharpness",
"makeup": "Clean editorial natural makeup",
"hair": "Sleek low bun with clean center part"
},
"environment": {
"location": "Contemporary art gallery",
"details": "White walls, track lights, light wood floor, framed abstract artwork"
}
}
Scene Pack 04 — Mur rose / casual lunettes (variation coiffure)
{
"task": "influencer_generation",
"subject_reference": "Use the person provided in the input portrait",
"scene_description": "A casual lifestyle photo outdoors in warm sunlight, sitting on a pastel pink ledge beneath a window with shutters. The subject looks to the side, relaxed and natural, like an unplanned friend photo.",
"outfit": "Beige plaid fitted t-shirt, simple grey mini skirt, brown leather knee-high boots, black round glasses, small green shoulder bag.",
"style": {
"vibe": "UGC premium, casual summer, real-life moment",
"color_palette": "Pastel pink, warm skin tones, beige and brown",
"look": "Natural light, mild shadows, non-commercial"
},
"composition": {
"shot_type": "Seated 3/4 body",
"framing": "Subject centered-left, bag visible, legs crossed, 3:4",
"pose": "Relaxed seated posture, hands resting naturally",
"camera_constraint": "Eye-level to slightly above seated position, no studio symmetry"
},
"technical": {
"lighting": "Direct warm sunlight with realistic shadow edges, avoid HDR",
"camera": "Shot on iPhone, vertical 3:4",
"makeup": "Natural minimal makeup",
"hair": "Wet hair down, natural clumped strands, realistic damp texture (intentional variation from the signature bun)"
},
"environment": {
"location": "Outdoor wall and window",
"details": "Pastel pink wall, white window frame, shuttered window"
}
}
Scene Pack 05 — Date night / restaurant (ton prompt)
Reprends exactement le prompt principal plus haut (c’est ta version “gold”).
Scene Pack 06 — Plage / marche au bord de l’eau
{
"task": "influencer_generation",
"subject_reference": "Use the person provided in the input portrait",
"scene_description": "A candid beach photo on wet sand near the shoreline. The subject walks slowly toward camera with a relaxed expression, ocean waves behind, morning or late-afternoon sun with long soft shadows. Realistic wind and light, non-commercial travel vibe.",
"outfit": "Two-tone bikini (black and cream) with a light beige sheer cover-up robe. Minimal gold jewelry.",
"style": {
"vibe": "Travel lifestyle, calm, elegant beach candid",
"color_palette": "Soft blues, beige, black, natural skin tones",
"look": "Real sunlight, mild haze, natural texture, no HDR"
},
"composition": {
"shot_type": "Full body portrait",
"framing": "Subject centered, horizon level, 3:4",
"pose": "Natural walking stride, arms relaxed",
"camera_constraint": "Horizon must stay level, realistic perspective, no extreme wide distortion"
},
"technical": {
"lighting": "Natural sun, realistic highlights, avoid HDR",
"camera": "Shot on iPhone, vertical 3:4",
"makeup": "Natural minimal makeup",
"hair": "Sleek low bun with clean center part"
},
"environment": {
"location": "Beach shoreline",
"details": "Ocean waves, wet sand reflections, clear sky"
},
"physics_override": {
"gravity_reference": "World gravity is vertical regardless of camera tilt"
}
}
Checklist anti-drift (très court)
- Réduis le nombre d’objets si ça part en vrille (1 verre + 1 bougie + 1 menu max).
- Fige 3 choses : (1) angle caméra, (2) focale iPhone, (3) coiffure signature.
- Ajoute une contrainte explicite “same identity as reference, same facial features, no change of face”.
- Itère par micro-changements (regard / main / placement objets) au lieu de tout changer d’un coup.
Conclusion
Avec cette structure, tu peux “builder” une influenceuse comme un personnage de série : une identité fixe, et des épisodes (scènes) qui changent.
Le plus important n’est pas la poésie du prompt : c’est la séparation invariants / variables + les contraintes caméra.
