Fellipe Sales
Prompt liberado
Você comentou, aqui está

O prompt de
30 segundos.
Um take só.

Aquele vídeo não foi montado com vários clipes colados na edição. É uma geração única, de ponta a ponta, com a fala, os movimentos de câmera e o que ela faz com as mãos cronometrados segundo a segundo dentro do próprio texto.

O prompt inteiro está nesta página. Copia, cola e roda.

30s num take contínuo Seedance 2.5 na Morphix Áudio e lip-sync na mesma geração Nenhuma edição depois
01

Cria sua conta na Morphix

A Morphix é onde os modelos de vídeo ficam reunidos num lugar só. Você entra, escolhe qual modelo quer usar e roda. Não precisa instalar nada, é tudo pelo navegador.

Cria a conta pelo botão abaixo e já entra direto na tela de geração.

Criar minha conta na Morphix Abre em outra aba. Volta aqui depois pra pegar o prompt.
02

Escolhe o Seedance 2.5

Esse é o detalhe que decide tudo. O Seedance 2.5 é o único modelo que gera 30 segundos numa tomada só. Os outros param em 8, 10 ou 15 segundos, e aí o prompt inteiro perde o sentido, porque ele foi escrito em cima de uma linha do tempo de trinta.

Dentro da Morphix, vai na aba Video, abre o seletor de modelo e marca Seedance 2.5. Depois deixa o resto exatamente assim:

Configuração da geração
Modelo
Seedance 2.5
Duração
30s
Proporção
16:9
Áudio
On
Resolução
720p
Start Frame / References
vazios

A duração não abre em 30 sozinha, você precisa mudar na mão. Se esquecer disso, o vídeo sai cortado no meio da fala.

03

Cola o prompt e gera

É longo de propósito. Cada bloco está fazendo um trabalho: travar que é uma tomada só, impedir que ela fique parada esperando a fala, descrever a mulher com detalhe suficiente pra ela não virar outra pessoa no meio, e marcar o tempo exato de cada frase e cada gesto.

Copia inteiro, sem tirar nada, e cola no campo de texto da Morphix.

Prompt completo

ONE UNBROKEN TAKE. Thirty seconds of continuous recording from a single camera that never stops rolling. There are no cuts, no edits, no transitions, no jumps, no angle changes, no reframing, no shot changes. Time never skips. This is one single continuous recording from beginning to end.

ABOUT THE TIMESTAMPS BELOW. The timestamps mark when things occur inside the one continuous take. They are not shot boundaries, not clip divisions, and not edit points. Nothing changes at a timestamp except what she is doing. The camera keeps rolling straight through every one of them.

NO DEAD AIR. She never stops, stares, or holds a blank expression between actions. Every movement overlaps the next, she is already beginning the following action before the current one finishes, and she talks through her movements rather than pausing to speak. There is no moment where she sits motionless waiting for a line, and no moment after a line where she holds a still expression.

NO VISIBLE PHONE OR DEVICE. The camera is her phone. No phone, screen, handheld device or recording equipment appears anywhere in frame at any point. During the opening she is carrying the camera, her right arm is extended toward the lens and heavily foreshortened at the frame edge, and her hand is out of frame. Her left hand is empty.

NO BRANDING. No logos, no brand names, no product labels, no readable packaging text anywhere in frame at any point. Every object in the scene is plain and unbranded.

THE WOMAN. She is the same person in every frame of these thirty seconds and her face must not drift, morph or change at any point. She is around thirty. Fair skin with a neutral, very slightly olive undertone. Dark brown, almost black-brown wavy hair gathered into a loose messy bun set high at the back of her head, with a few strands escaping at the temples and at the nape, slight frizz, clearly not styled. A narrow face with defined cheekbones and slightly hollow cheeks, a long slim neck, visible collarbones, a slender build. Green-hazel eyes, slightly hooded, with a faintly tired set to them. Thick natural dark brown eyebrows, uneven, never plucked into a shape. A narrow straight nose with a very slight bump at the bridge. Full lips in a natural pale rose, no lipstick. Small ears, the near one partly showing through loose hair. She wears no makeup.

Her freckles are the strongest identifying mark and must stay consistent throughout: scattered light brown freckles across both cheeks and over the bridge of the nose, becoming denser down the sides of the neck and across the shoulders and upper chest, plus two or three small dark moles on the neck and one on the near shoulder.

She wears a muted sage green ribbed cotton tank top with thin straps, an ordinary at-home piece, slightly loose at the neckline. She wears tortoiseshell prescription glasses, rounded rectangular frames in warm reddish-brown acetate, until she removes them at 14.4s, after which she is bare-faced for the rest of the take and the glasses never reappear.

FORMAT. Horizontal 16:9 landscape. Front-facing phone camera. Mild wide-angle distortion.

CAMERA GEOMETRY. A mirror stands on the desk; she sits facing it. The camera rests on the desk to her left, offset roughly 35 degrees from her facing direction and angled back toward her. It is never in front of her. Her resting position is three-quarter, left side nearer the lens. Her natural forward gaze, into the mirror, exits toward frame-left. To address the viewer she turns her head to her right toward the lens, and that turn is always a visible deliberate movement. The mirror sits just past the left frame edge; only its dark wooden edge and a soft out-of-focus sliver of her shoulder and hair are visible at the extreme border. Her face is never readable in it. Frame-right is open room: balcony doors, sheer curtains, plants. A soft leather handbag sits on the desk in the lower left of frame.

LIGHTING. Overcast-soft daylight through the balcony doors from frame-right, diffused by sheer curtains, the only meaningful source. Cool-neutral, roughly 5600K. Paper lantern off, bedside lamp off. Soft open shadows, heavy fill off the pale walls. Because the light comes from frame-right and she sits three-quarter, it rakes across her cheekbone and jaw, the side nearer the lens falls into soft shadow, the far side is lit.

DELIVERY. Light Australian accent. Relaxed conversational pace, roughly 3.4 words per second. Lines run into one another naturally without gaps.

DIALOGUE, LOCKED VERBATIM. She speaks exactly these words, in exactly this order. No improvisation, no rewording, no reordering, no additions, no omissions. This is the complete and only dialogue in the video.

LINE 1 (0.3s-4.0s) - "Ok guys, I had to show you how insane AI has become lately."

LINE 2 (4.8s-7.9s) - "Like, I'm not real. None of this is."

LINE 3 (8.6s-13.6s) - "But this isn't the usual AI workflow where you generate multiple videos and edit them together."

LINE 4 (14.2s-18.3s) - "I used one, extremely long but detailed prompt, and got a 30 second clip."

LINE 5 (18.6s-24.5s) - "It includes the dialogue, camera and body movements, with exact timestamps, so you control every second of it."

LINE 6 (24.8s-29.4s) - "And if you want the prompt, I'll send it to you, just comment 'AI'."

CONTINUOUS BEHAVIOUR. She is never still or neutral. Eyebrows move constantly with her speech, lifting on emphasis. She blinks at irregular intervals, sometimes twice in quick succession. Her mouth stays active between words, small lip presses, a slight purse while thinking. Her head makes small tilts and nods tracking her own sentences. Shoulders shift and settle. She adjusts her seated weight two or three times. Her hands are almost never at rest; when not gesturing they touch the desk edge, her own forearm, a strand of hair. Her eyeline moves continuously between three places, the lens, the mirror off frame-left, and whatever her hands are doing. She takes audible breaths between lines. She never freezes to deliver a line.

THE RECORDING, IN ORDER.

(0.0s-4.0s) The recording is already running when it reaches us. She is mid-stride, walking in from the hallway, the camera carried low around hip height and tilted upward so it catches her jaw, the unlit paper lantern and the ceiling. Her right arm extends toward the lens, foreshortened and cut by the frame edge; her hand is not visible. Her left hand pushes hair off her face as she walks. She is looking ahead into the room, not at the lens. The frame bounces softly with each step and drifts and self-corrects the way a carried camera does. At 0.8s exposure re-levels as she comes out of the dim hallway into the daylight, the frame brightens visibly and the white balance cools. Rolling-shutter jello runs through this whole stretch. LINE 1 runs from 0.3s to 4.0s, slightly breathy from walking, the last few words landing as she is already reaching toward the desk.

(4.0s-4.8s) The camera is lowered onto the desk to her left and turned a few degrees back toward where she is about to sit. The frame swings, tips, and settles with two decaying wobbles between 4.2s and 4.5s. Fingers pass briefly across the bottom edge of the glass at 4.3s. At 4.6s a fingertip nudges the angle once, correcting it, then withdraws. Autofocus hunts and locks by 4.8s as she drops into the chair. The camera now remains in this exact position and angle, recording without interruption, from 4.8s to 29.5s. Every change in the image across those twenty-five seconds comes from her movement, never from the camera.

(4.8s-8.0s) She is already speaking LINE 2 as she sits, the line begins at 4.8s while she is still lowering into the chair, and she turns her head to her right to find the lens on the way down, so that the sit, the turn and the first words all happen together rather than in sequence. One hand comes up loosely around 5.6s, palm turning slightly outward on "none of this is" at 6.6s, and her shoulders lift in a small half-shrug. The tone is flat and matter-of-fact and breaks into a short quiet giggle at 7.4s, eyebrows lifting, a real amused smile arriving late. She stays at comfortable conversational distance, head and shoulders in the centre of the wide frame with room visible on both sides, and she does not lean in close. Skin detail is at maximum through this stretch under the raking daylight, which hides nothing: visible pores across the nose and inner cheeks, fine peach fuzz along the jaw and upper lip catching the side light, natural oil sheen on the forehead and nose bridge, faint capillary redness under the eyes, individual eyebrow hairs separated, moisture on the lower lash line, translucency through the earlobe nearer the window. Autofocus breathes once around 5.2s and settles. At 7.9s the giggle is still finishing as her hand is already dropping toward frame-left.

(8.0s-13.6s) Her hand finds the soft leather handbag at 8.0s and draws it across the desk toward her, the bag moving with weight, the leather flexing and the strap sliding off the desk edge and swinging. LINE 3 starts at 8.6s, before the bag has finished moving. She talks while looking down into it, both hands working through it, feeling rather than looking, and her head comes up to the lens at 9.8s and again at 11.9s before dropping back down. At 11.2s one hand lifts out of the bag and makes a small stacking motion, palm down, two short beats, then goes straight back in. At 12.7s she blows a loose strand off her face with a short upward puff without stopping talking.

(13.6s-18.3s) At 13.6s she finds a plain unbranded foil sachet and draws it out. The blank mirrored foil carries no text, no logo and no label, and picks up the daylight so a soft highlight travels across it as it clears the bag. She does not present it to the camera, she sets it flat on the desk beside her at 14.0s and keeps talking. LINE 4 starts at 14.2s. Both hands come up and she removes the tortoiseshell glasses in one unhurried movement between 14.4s and 15.3s, folds them one-handed and sets them down, and as she does this she turns to the mirror off frame-left, the automatic thing anyone does taking glasses off in front of one. From 15.3s to 17.4s we see her go three-quarter, then near-profile, checking her own face. Faint pressure marks remain on the nose bridge; she blinks three times adjusting and at 16.6s rubs the bridge of her nose with two fingers, still looking at her reflection. At 17.4s she turns back to the lens for the last few words of the line with a small emphatic nod and raised eyebrows.

(18.3s-24.5s) At 18.3s she picks the sachet back up, grips the top notch and tears it open in one steady pull between 18.9s and 19.8s. The foil separates along a jagged edge, the torn strip bends and stays attached, and the packet visibly softens as the contents shift, with wet glisten at the opening. Her eyes drop to it for the tear and come back up by 20.0s. LINE 5 runs from 18.6s to 24.5s, gestured with the open sachet still in one hand, three small separations in the air at 20.4s, 20.9s and 21.4s, low and close to her body, not raised or presentational, while her free hand tucks a front strand behind her ear at 22.0s. A quick raised-eyebrow beat and a small closed-mouth smile arrive at 23.9s on the last few words.

(24.5s-29.4s) At 24.5s she pulls a folded white sheet mask out. It is heavy with serum, sagging under its own weight, sticking to itself, dripping once onto the desk at 24.9s. She makes a small involuntary face at the wetness while still talking, then unfolds it with both hands between 25.2s and 25.9s, the fabric stretching and going semi-translucent where saturated. LINE 6 starts at 24.8s and runs underneath all of this. At 25.9s she turns to the mirror to put it on, and from 25.9s to 28.9s she is in near-profile to the lens, hands up at her face, working on her own reflection off frame-left. She aligns the eye and mouth cutouts using the mirror and presses the mask on from the forehead down starting at 26.4s. Air pockets form at the cheeks and she smooths them with her fingertips at 27.3s; an edge lifts near the jaw at 27.9s and she presses it back; the fabric conforms over the nose bridge. Skin tone and freckles read through the wet material, and the serum sheen catches the daylight from frame-right, throwing a strong specular edge along her cheekbone in profile. Her voice changes slightly from 26.4s as the mask goes on, marginally more constrained around the mouth, marginally more distant while she is turned away.

(28.9s-29.5s) She gives her reflection a final check and at 28.9s turns her head back to her right to find the lens again, the same turn as at 4.8s, closing the loop. Her hands come down and settle. She holds still, mask on, looking directly at us, the stillness after five seconds of hand activity being what makes the ending land. This is the only stillness in the entire take. One small gesture toward the lens on "comment 'AI'" at 29.0s, and the hand returns to the desk. Her eyes are visible through the cutouts and the smile is readable in them and in the movement of the fabric at her cheeks.

(29.5s-30.0s) She reaches forward. Her hand enters frame, going soft and dark as it nears the glass, and the frame tips as the camera is lifted off the desk. The recording stops there, mid-motion, still running until the instant it ends.
o passo depois desse

Esse prompt faz uma cena. Com uma mulher que o texto inventou.

Repara numa coisa: toda vez que você roda, nasce outra pessoa. Parecida, nunca a mesma. Dá pra postar um vídeo assim. Não dá pra montar um perfil, porque no terceiro post ninguém reconhece mais quem é.

O Influencer AI resolve os dois lados. Primeiro um prompt entrevista você e monta o personagem fixo, com rosto, corpo, cabelo e roupa escritos de um jeito que não deriva. Depois você tem 300 cenas prontas pra colocar ele dentro.

  • o prompt que entrevista você em 14 perguntas e monta o seu avatar
  • 300 cenas prontas em 10 pilares, com busca e filtro
  • cada botão copia o prompt inteiro já com o seu personagem dentro
  • as 7 regras que tiram a cara de IA e fazem passar por foto de gente
acesso completoR$99 Quero criar o meu acesso imediato por e-mail, pra sempre

Quatro coisas que eu descobri apanhando

Gastei crédito à toa pra aprender isso. Você não precisa.

Não coloca foto de referência

O Seedance recusa qualquer imagem com rosto humano realista, seja no References ou no Start Frame. É trava anti-deepfake e não tem como contornar. Deixa os dois vazios. A identidade dela vem do bloco de descrição que está dentro do prompt.

A fala fica em inglês

Testei em português e o lip-sync desmonta nas nasais, aquele "não", "edição", "gente". Português tem vogal nasal e o modelo foi treinado em inglês, onde isso quase não existe. Deixa a fala em inglês e legenda depois.

Cada rodada vem uma pessoa diferente

Sem foto de referência, o rosto nasce do texto. Rodadas diferentes dão mulheres parecidas, não idênticas. Escolhe uma tomada e fica com ela, porque não dá pra refazer a mesma com o rosto igual.

Os tempos são um alvo, não um relógio

O modelo mira nos timestamps mas não crava no milissegundo. Os gestos chegam uns décimos antes ou depois. É o suficiente pra coreografia funcionar, e é isso que faz parecer uma pessoa gravando, não um robô cumprindo marcação.