Le prompt de l'influenceuse IA (MakeUGC + Claude + GPT Image 2 + Seedance 2.5)

Le prompt complet qui cartographie seconde par seconde la vidéo de ton influenceuse IA, plus les trois outils à brancher.


Les trois outils à brancher

ÉtapeOutilCe qu'il fait
1MakeUGC, connecté à Claude (Connect to Claude dans API Integration)tu génères tous tes visuels en parlant à Claude
2GPT Image 2il fabrique ton avatar et l'image produit
3Seedance 2.5il transforme les images en vidéo

Les deux demandes à faire dans Claude, mot pour mot :

build an avatar for UGC videos using GPT Image 2
create the video using the avatar and the product using Seedance 2.5

Le prompt, section par section

Tu colles tes deux images (<<image_1>> ton avatar, <<image_2>> ton produit), puis ce bloc :

== REFERENCE MAP ==
<<image_1>> (tongue-out selfie in yellow hoodie) MASTER REFERENCE. Character identity, face, skin, hair, makeup, jewelry, room, wall, lighting, color temperature, voice, delivery, overall style and vibe. Everything defaults to this image.
<<image_2>> (flat-lay hoodie) PRODUCT. Butter-yellow oversized sherpa-lined hoodie. Exact color, exact cream sherpa hood lining, kangaroo pocket, dropped shoulders, ribbed cuffs and hem.

== CAPTURE + AUDIO SPEC ==
iPhone 17 Pro Max, 24mm main lens, vertical 9:16, handheld-grade sharpening, mild shadow noise, HDR highlight rolloff. No grade, no LUT, no cinematic polish. Light exactly as <<image_1>>: flat soft daylight from off-frame front-left, warm-white wall, matte black horizontal wall fixture behind her at ear height, small round ceiling vent top of frame. No fill, no bounce, no rim. Phone is propped on a dresser for Shots 1, 2, 4, 5 - never visible in frame. Shot 3 is the same phone in her hand, rear camera. Audio: single phone mic at room distance. Thin midrange, auto-gain pumping as she moves nearer and further, real reflections off hard walls, clothing rustle, loud close fabric handling in Shot 2, muffled pass in Shot 4. No denoise, no compression, no music, no room tone bed. Voice matched to <<image_1>> - same speaker, breathy conversational American, mid-pace. Include breath intakes, vocal fry on line ends, one mouth click, one micro-stumble. No captions, no text overlay, no watermark. The only text in frame is the woven neck label.

PART 1 - SHOT BREAKDOWN (15s, five shots)

Le découpage des cinq plans

Le dialogue

== DIALOGUE (~33 words, gaps intentional) == et la première ligne du plan 1, à 13 s : Okay. I just got this and I'm obsessed. Tu écris les 33 mots, et tu laisses les silences.

Une question ? Réponds directement à ce message, je lis tout.