Haz tu mini yo con IA y métela en tu vídeo.

Character model sheet for a tiny animated sidekick: a chibi cartoon version of the young woman in the reference photos, who will pop up next to her in a social video as her cheeky inner voice. Proportions: big head, about one third of total body height, small body, tiny hands and feet, like a premium chibi mascot. Art style: polished modern 2D animation, the look of a high-budget animated commercial: smooth variable-weight brush outlines in warm dark brown (not black), soft gouache-like shading with a gentle rim light, glossy hair shapes with two highlight streaks, big sparkling eyes with a pink catchlight, subtle blush. Stylish and cute, never generic emoji or avatar style. Keep her recognisable: [long straight light-brown hair with warm auburn tones and a centre parting, fair skin, rosy cheeks, light brown almond eyes, full soft lips, thin gold necklace]. Outfit: [fitted white tank top, light blue high-waisted jeans, soft pink sneakers]. Layout on a plain warm off-white paper background: top row, full body front view, three-quarter view, side view, and one action pose waving hello; bottom row, six head expressions: talking with mouth open, cheeky smirk, surprised, unimpressed with one eyebrow up, laughing with eyes closed, winking. No text labels. Identical character design in every drawing.

Character model sheet of the SAME chibi character as in the first reference image (the approved model sheet): exactly the same art style, proportions (big head about one third of the height), face, big brown eyes with pink catchlight, [long straight light-brown hair with centre parting], warm brown outlines and soft gouache shading. Change ONLY her outfit and add glasses, copying them from the photos of the real woman: [black oval glasses with thin frames; a white cropped tank top with thin spaghetti straps and a small black Nike swoosh on the chest; pink high-waisted bike shorts; white sneakers; thin gold necklace]. Layout on a plain warm off-white paper background: top row, full body front view, three-quarter view, side view, and sitting down with legs dangling as if sitting on someone's shoulder; bottom row, six head expressions, all wearing the glasses: talking with mouth open, cheeky smirk, surprised, unimpressed with one eyebrow up, laughing with eyes closed, winking. No text labels. Identical character in every drawing.

Draw exactly the character from the first reference image (front view, top-left of the sheet): same face, [black oval glasses], hair, [white cropped tank with small black swoosh, pink bike shorts, white sneakers, gold necklace], same art style, outlines and shading. Single character, full body, standing facing the viewer with a friendly closed-mouth smile, arms relaxed, centred with space around. Background: a completely flat uniform bright green (#00FF00), no ground shadow, thin white outline around the character.

Edit this exact image. [Shrink the beauty mole below the lower lip to a TINY dot: about one third of its current diameter, soft light warm brown, only noticed when looking closely. Same position.] Do NOT touch the glasses: keep the same thin round black glasses exactly as they are. Change NOTHING else: same pose, size, face, eyes, hair, outfit, outline and the same flat bright green background.

ffmpeg -i su-frase.m4a -af "aresample=48000,asetrate=48000*1.2599,aresample=48000,atempo=0.7937" su-frase-mini.wav

Edit the reference video. Keep the woman, her face, her lip movements, her hands, the room, the light and the camera exactly as they are. Add one small 2D cartoon girl: the character from the reference images ([long straight light-brown hair, thin round black glasses, white tank top with a small black swoosh, pink bike shorts, white sneakers, a tiny beauty mark under her lower lip]), drawn in the same soft chibi illustration style, with no white sticker outline. She is about as tall as the woman's head and sits on the woman's right shoulder: the shoulder closest to the LEFT EDGE of the frame, NOT the shoulder on the right side. Legs dangling over the front of the shoulder, with a soft contact shadow on the shoulder and the same warm room light on her. She is there from the very first frame, starts seated on that shoulder and moves with it until the action below explicitly changes her position. If no change of position is specified, she stays seated there for the whole clip. Action in order: [from 0 to 2 seconds she looks at the camera, smiling proudly, swinging her feet and blinking. At about 2 seconds, when the woman says she already regrets it, the cartoon girl turns her head to look at the woman's face, offended, and crosses her arms. From 3.4 to 4.2 seconds she speaks the line in the reference audio, '[su frase]', her mouth moving in sync with that voice, looking at the woman sideways with a smug narrowed-eyes face. Until the end she keeps the smug look and raises one eyebrow.] Real cartoon animation: head turns, blinks, squash and stretch. No other characters, no text, no extra objects, no change to the woman.

Action in order: from 0 to [2.6] seconds, while the woman talks, the cartoon girl listens proudly, chin up, chest out, nodding and smiling, blinking. From [2.75] to [5.2] seconds she speaks the line in the reference audio, '[Y la primera versión era un emoji con pelo]', her mouth moving in sync with that voice, with a disgusted, cringing face, pointing at her own head and hair with both hands. Until the end she keeps the cringing face.

Action in order: from 0 to [1.8] seconds, while the woman talks, the cartoon girl nods along, curious, blinking. From [1.95] to [3.45] seconds she speaks the line in the reference audio, '[Y salí yo. Guapísima.]', her mouth moving in sync with that voice, flipping her long hair back with one hand and striking a proud, glamorous pose with her other hand under her chin, eyes half closed, very pleased with herself. Until the end she keeps the proud pose.

Action in order: from 0 to [1.1] seconds, while the woman talks about her mole, the cartoon girl touches the tiny beauty mark under her own lower lip with one finger. From [1.27] to [3.35] seconds she speaks the line in the reference audio, '[Del tamaño de una aceituna]', her mouth moving in sync with that voice, with a horrified wide-eyed face, holding her thumb and index finger apart to show the size of an olive. From [3.4] seconds to the end, while the woman talks again, she lets out a relieved sigh with one hand on her chest.

Action in order: from 0 to [2.6] seconds, while the woman talks, the cartoon girl listens with her arms crossed, skeptical, one eyebrow raised. From [2.81] to [4.01] seconds she speaks the line in the reference audio, '[Y parecía una pegatina]', her mouth moving in sync with that voice, and for a moment she freezes completely stiff, arms straight down at her sides, staring blankly forward like a stiff paper doll, then relaxes and shrugs with a deadpan face.

Action in order: from 0 to [0.6] seconds she sits on the woman's right shoulder: the shoulder closest to the LEFT EDGE of the frame, NOT the shoulder on the right side of the frame, legs dangling. At [0.6] seconds the woman raises her open hand, palm up, on the left side of the frame at chest height, just below that same shoulder: the cartoon girl hops straight down from the shoulder and lands standing on that open palm by [1.3] seconds, and stays standing on the palm, moving with the hand, with a soft contact shadow on the palm. From [2.03] to [3.93] seconds she speaks the line in the reference audio, '[Y aquí estoy. Viva. Y opinando.]', her mouth moving in sync with that voice, arms spread wide, proud and theatrical. At [3.8] seconds the woman swings her arm up and throws her: the cartoon girl is launched upward, flies up and out of the top of the frame, spinning, and is gone by [4.3] seconds. After that the cartoon girl does not appear again.

Monta y termina mi vídeo de mini yo usando los archivos de [CARPETA DEL VÍDEO]. Ejecuta el trabajo sobre los archivos y entrega un MP4 final, no solo instrucciones para que yo edite.

MATERIAL Y PREPARACIÓN
1. Haz un inventario de vídeos, audios, referencias del personaje, guion y subtítulos. Identifica cada plano y cada frase. Si una correspondencia no está clara, pregunta; no inventes diálogos ni tomas.
2. Comprueba que puedes leer y procesar los archivos y que tienes FFmpeg y las herramientas de imagen, máscaras y seguimiento que necesites. Indica cualquier acceso o dependencia que falte. No presupongas que un conector o el control del navegador está disponible.
3. Trabaja en una copia nueva. Conserva los originales y cualquier versión aprobada. Si existe un montaje aprobado, mantén su orden, duración, voz, subtítulos y gráficos salvo la corrección pedida.
4. Si parto de brutos, construye el montaje siguiendo el guion. Prepara un WAV por plano, con silencio fuera de la frase de la mini y la misma duración que el plano. Usa mi voz en español, sin traducirla ni sustituirla por una voz generada. Si ya está procesada, no vuelvas a subir el tono.
5. Si la voz de la mini aún está sin procesar, sube cuatro semitonos conservando la velocidad con este filtro: aresample=48000,asetrate=48000*1.2599,aresample=48000,atempo=0.7937. No añadas eco. Comprueba duración y sonido.

MONTAJE Y ACABADO
1. Conserva mi vídeo y mi audio originales: Pon mi grabación como capa inferior y bloquea sus pistas de audio. Silencia el audio de los vídeos generados. Así conservas mi cara, mi voz y el diálogo en español durante todo el montaje.
2. Integra solo la mini: Encima del original, coloca la mini del vídeo generado. Aíslala con una máscara y seguimiento de movimiento, sin incluir mi cara ni mi pelo real. Si se mueve de sitio, ajusta la máscara con fotogramas clave. Mantén una escala coherente y una sombra suave de contacto con el hombro o la mano. Resuélvelo con herramientas de máscaras y seguimiento, por código o mediante un editor disponible.
3. Sincroniza la boca con mi WAV: Si la boca no coincide con la frase, usa Lipsync en Higgsfield y selecciona Sync Lipsync 3. Sube un recorte del vídeo de ese plano en el que solo aparezca la cara de la mini, sin mi cara real, y el WAV del mismo plano con sus silencios. Conserva todo el movimiento y la duración del plano. Usa el resultado como corrección visual; la pista de audio sigue siendo mi grabación original.
4. Aplica la corrección solo a la boca: Alinea el resultado de Sync con el mismo comienzo del plano. Coloca sobre la mini únicamente la zona inferior de su cara, con una máscara de borde suave que siga su movimiento. Conserva ojos, gafas, pelo y cuerpo del plano base. Si la salida cambia de duración, alinéala antes de componer; no desplaces la voz para compensarla.
5. Cierra la boca durante los silencios: Revisa el inicio, las pausas y el final de cada frase. Si hay movimiento de labios sin voz, toma una boca cerrada de la misma mini, con el mismo ángulo e iluminación, y aplícala solo durante el silencio con una máscara suave y seguimiento. Si el rostro gira, usa una referencia correspondiente a ese ángulo para evitar una cara pegada.
6. Unifica el pelo y el contorno: Compara la mini al mismo tamaño en todos los planos. El contorno marrón debe mantener un grosor visual coherente. Para una muesca o un temblor, usa un borde limpio de la misma mini y sigue el movimiento de cabeza y cuerpo por separado. Limita la máscara al pelo: no debe invadir mejillas, gafas o manos. Recupera el fondo exterior desde mi vídeo original.
7. Comprueba el archivo exportado: Exporta en vertical a 1080 × 1920, con los fps de mi montaje y audio a 48 kHz. Mira el archivo completo con sonido a velocidad normal; después revisa de cerca boca, silencios, pelo, contorno y cambios de plano. Comprueba que solo suene mi voz, que las respuestas estén completas y que no haya verde, halos blancos, dobles bordes ni movimientos en las pausas.

GENERACIONES Y ACCESOS
Reutiliza primero los clips y referencias existentes. Si hace falta Sync Lipsync 3, prepara el recorte de la mini y su WAV exacto. Si tienes acceso autorizado a Higgsfield, realiza la operación; si no, deja esos archivos preparados y dime la operación concreta que requiere mi intervención. Antes de gastar créditos adicionales, indica modelo, finalidad y coste mostrado y pide autorización. No publiques ni compres suscripciones.

VERIFICACIÓN Y ENTREGA
Comprueba el archivo exportado, no solo los intermedios. Verifica formato, duración, fps y audio; revisa todos los planos, los silencios y los empalmes. Mantén el audio aprobado idéntico cuando la reparación sea solo visual. No des por buena una boca que habla sin voz ni un parche que invade la mejilla.
Entrega el MP4 final en una copia nueva, los archivos editables o scripts para reproducir el montaje y un resumen breve de qué has comprobado. Distingue las comprobaciones automáticas de la revisión visual y de sonido realmente realizada. Si no puedes comprobar algo, indícalo. Abre el vídeo final para que lo revise.