AI 티를 없애는 실사 영상 프롬프트 설계법: Opus 5.5 디렉팅과 Seedance 2.5 레퍼런스 맵

생성형 AI 비디오 모델이 비약적으로 발전했음에도 불구하고, 대다수의 AI 영상은 인위적인 피부 질감, 부자연스러운 조명 변화, 렌즈 특성의 부재로 인해 특유의 '가짜 티'를 완전히 벗겨내지 못하는 한계를 보입니다. AI 크리에이터이자 프롬프트 엔지니어인 @abxxai(Abdul Shakoor)는 최신 추론 모델인 Claude Opus 5.5를 연출 총괄 디렉터로 세우고, 이미지·비디오 생성 모델(Seedance 2.5 등)에 정교한 물리적 결함과 카메라 광학 특성을 주입하여 핸드헬드 실사 촬영물과 구별하기 어려운 15초 원테이크 영상을 제작하는 4단계 프롬프트 아키텍처를 공개했습니다.

이미지 출처: @abxxai (Abdul Shakoor)

이번 팁의 핵심 철학은 "완벽함(Cinematic)을 요구하지 말고 결함(Flaws)을 지시하라"는 것입니다. 모공, 솜털, 필름 그레인, 흔들리는 핸드헬드 동선, 초점을 잡는 포커스 헌팅(Focus hunt), 그리고 광원 자체의 변화가 아닌 카메라 센서의 자동 노출(Auto-exposure) 적응을 프롬프트에 명시적으로 설계하는 실전 가이드입니다.

1단계: 톤앤매너와 광학 결함을 정의하는 단일 스틸 컷 프롬프트

실사 영상 제작의 첫 단추는 영상 전체의 룩(Look), 카메라 하드웨어, 광원 방향, 의도된 광학 결함을 고정하는 기준 스틸 이미지를 생성하는 것입니다. 단순히 '빈티지한 요트' 같은 모호한 수식어를 피하고, 2000년대 초반 CCD 컴팩트 디지털카메라의 거친 플래시 특성과 하이라이트 번짐(Bloom/Halation)을 구체적인 수치와 물리 현상으로 기술합니다.

Two women framed by the open companionway of a small wooden sailing boat — one in the near foreground at frame right inside the cabin in a white ribbed tank top and wide cream linen trousers, seated on the varnished steps with one knee up, looking at the camera; the second up on deck through the hatch in a lilac knit polo and white trousers, holding the rigging with one hand and leaning out, facing the camera. Setting: a classic teak sailboat cabin, varnished wood panelling, a brass porthole, rope coils and a folded striped towel in the foreground, the open hatch revealing the sunlit deck, white sails, a stone harbour wall with small moored boats and flat blue sea under a pale sky, soft late-afternoon light. Shot on an early-2000s point-and-shoot digicam with harsh direct on-camera fill flash that aggressively illuminates the near woman and the cabin wood, flattening her features and putting sharp specular highlights on the varnish, the brass and her shoulders; the deck and harbour beyond stay visible but hazy, milky and underexposed relative to the flash. Warm, slightly saturated CCD tones, strong bloom and halation at the hatch edges, minor digital noise. Candid amateur, Y2K editorial, unpolished but stylized. 4:5.

프롬프트 핵심 포인트

• 카메라 및 센서 명시: 2000년대 초반 포인트앤슛 디지캠의 CCD 센서 색감 및 디지털 노이즈 정의.
• 직접 플래시 조명(On-camera fill flash): 근경 인물과 원경 해항 간의 노출 격차, 바니시 칠한 목재 표면과 쇄골에 맺히는 날카로운 반사광(Specular highlights)을 강제.
• 아마추어 스냅 감성: 인위적인 보정 대신 해치 경계부의 강한 블룸과 헤일레이션을 배치해 사실적인 스냅 룩 구축.

2단계: 캐릭터 얼굴·신체 일관성을 위한 16:9 레퍼런스 시트 고정

영상을 움직이기 전에 가장 중요한 과정은 인물의 얼굴이 재생 중 일그러지거나 표류(Face drift)하지 않도록 배우별 캐릭터 시트를 단독 16:9 패널로 고정하는 것입니다. 전신 4면도, 두상 3면도, 4가지 핵심 감정 표정 스트립, 의상 소재 및 장신구 클로즈업을 단일 프레임에 담아 모델에게 제공합니다.

캐릭터 A (선실 내부 인물) 레퍼런스 프롬프트

Photorealistic character reference sheet of one young woman, 16:9, plain light-grey seamless background, soft even daylight, no text, no labels. Top row: full body front, three-quarter, side profile, back view, standing relaxed and barefoot. Bottom row: close-up head front, three-quarter, profile, then a four-face expression strip — neutral, soft closed-mouth smile, mid-word talking, easy laugh. Small insets: layered gold necklaces, hands, linen fabric texture. She is in her early-to-mid 20s with sun-kissed light olive-tan skin, dense natural freckles across her nose and cheeks, warm flushed cheeks, hazel-green almond eyes, defined natural brows, a straight nose, and full lips with a soft rosy-nude tint. Her hair is very long, sun-streaked honey-brown with lighter blonde ends, center-parted, with face-framing pieces and a slightly salty, undone texture. Wardrobe: a white fine-ribbed scoop-neck tank top, cream loose-weave linen wide-leg trousers, and two thin gold chains — one with a tiny cross, one with a small oval medallion. Skin realism: visible pores, fine texture, faint unevenness, natural sheen on the cheekbones, no smoothing, no retouching, no beauty filter. Identical face, hair and body across every panel.

캐릭터 B (갑판 위 인물) 레퍼런스 프롬프트

Photorealistic character reference sheet of one young woman, 16:9, plain light-grey seamless background, soft even daylight, no text, no labels. Top row: full body front, three-quarter, side profile, back view, standing relaxed and barefoot. Bottom row: close-up head front, three-quarter, profile, then a four-face expression strip — neutral, knowing half-smile, mid-word talking, playful wink with a small head tilt. Small insets: knit texture of the polo, hands, trouser waistband. She is in her early-to-mid 20s with light warm-toned skin and a faint summer glow, light brown eyes, straight soft brows, a softly rounded face, and full lips in a natural nude. Her hair is long and loosely wavy, light brown with sun-bleached ends, center-parted and wind-tousled. Wardrobe: a lilac/lavender fine-ribbed short-sleeve knit polo with a three-button placket left open, and white high-waisted wide-leg trousers. Skin realism: visible pores, fine texture, faint unevenness, natural sheen, no smoothing, no retouching, no beauty filter. Identical face, hair and body across every panel.

• 피부 사실성(Skin Realism) 철칙: 모공(Pores), 잔털, 자연스러운 굴곡, 광대뼈의 자연스러운 유분을 명시하고, 피부 보정이나 뷰티 필터(No smoothing, no retouching, no beauty filter)를 전면 금지합니다.

3단계: 6분할 로케이션 레퍼런스 시트로 공간 연속성 확보

카메라가 이동하거나 시점이 달라질 때 배경 구조물이 왜곡되는 현상을 막기 위해, 요트와 주변 환경의 6개 앵글을 하나의 16:9 시트에 사전 정의합니다.

Photorealistic location reference sheet, 16:9 grid of six panels, no people, no text. Panel 1: wide exterior of a classic 1960s teak sailing yacht moored in a small Mediterranean stone harbour at golden hour — glossy varnished hull, white sail half raised, stone harbour wall, small white fishing boats, flat blue water. Panel 2 (hero angle): inside the cabin at seated chest height, looking up the varnished teak companionway steps to the open hatch — deep glossy mahogany panelling, a large brass porthole on the right wall, a small brass bell above it, a thick coiled cream-and-navy rope at left, a navy-and-white striped towel folded on a bench in the foreground, a bright slice of deck, rigging, sail and harbour through the hatch. Panel 3: on deck looking back toward the hatch — teak deck planks, cleats, winches, coiled lines, the hatch opening, harbour and low sun behind. Panel 4: close detail of the brass porthole, bell and varnish reflections. Panel 5: companionway steps close-up — worn varnish, a brass drawer pull, a small brass handrail. Panel 6: the harbour view from deck — stone wall, moored boats, flat sea, a pale golden sky with a low sun. Lighting across all panels: warm late-afternoon sun, glossy specular highlights on the varnish and brass, a darker cabin interior versus a bright milky exterior. Realism: real wear on the wood, salt marks, scuffed brass, slightly frayed rope.

• 세부 마모도와 텍스처: 목재의 마모 흔적, 소금기 자국, 긁힌 황동 표면, 풀린 밧줄 가닥 등을 명시하여 현실적인 질감을 보장합니다.

4단계: Seedance 2.5 15초 무편집 원테이크 비디오 디렉팅 프롬프트

준비된 4장의 레퍼런스 이미지를 바탕으로, Opus 5.5가 작성한 타임라인별 연출 프롬프트를 Seedance 2.5 비디오 생성기에 입력합니다. 컷 전환 없이 하나의 카메라가 선실 내부에서 갑판 위로 이동하는 15초 연속 롱테이크(Continuous Handheld Take) 프롬프트입니다.

=== REFERENCE MAP ===
<<<image_1>>> (boat cabin) → exact first frame: woman in white ribbed tank seated low beside the varnished companionway steps at frame right; woman in lilac polo up on deck holding the rigging with her right hand; white sail, harbour wall and low sun beyond. Sets cabin, light, framing and camera look for the whole take.
<<<image_2>>> (cabin girl sheet) → <<<CHAR_A>>> face, freckles, hair, body, jewellery, wardrobe.
<<<image_3>>> (lilac girl sheet) → <<<CHAR_B>>> face, eyes, hair, body, wardrobe. She is the lilac-polo woman in <<<image_1>>>; in her close-up her face follows <<<image_3>>> exactly.
<<<image_4>>> (location sheet) → the teak yacht: mahogany cabin, companionway with brass handrail on the right, brass porthole and bell on the right wall, cream-and-navy rope coil and navy-striped towel at left, teak deck with chrome winches, stone harbour wall with small lighthouse, moored white boats, old stone houses, low golden sun. Environment only.

=== CHARACTERS ===
<<<CHAR_A>>> → early-to-mid 20s, sun-kissed light olive-tan skin, dense freckles on nose, cheeks and shoulders, flushed cheeks, warm hazel-brown eyes, very long sun-streaked honey-brown hair with blonde ends, center part. White fine-ribbed scoop tank, cream loose-weave linen wide-leg trousers, two thin gold chains (tiny cross, small oval medallion). Barefoot. Right hand = talking hand, left hand = climbing hand; never swap.
<<<CHAR_B>>> → early-to-mid 20s, light warm skin, faint freckles, grey-green eyes, small gold hoops, thin gold chain, long loosely wavy light-brown hair with sun-bleached ends. Lilac ribbed short-sleeve knit polo with open cream-buttoned placket, white high-waisted pleated wide-leg trousers. Barefoot.
Neither woman touches the camera. The person filming is never seen and never speaks; their only sound is one soft wordless breath of a laugh.

=== EMOTIONAL CORE ===
<<<CHAR_A>>> is not selling. She is letting one close friend in on something she loves and is amazed nobody else noticed. Soft eyes, a smile that starts in the eyes first, a quiet thrill on "shots like this," real tenderness naming the skin, the light, the grain. On "the camera shakes" she laughs for real, caught off guard — the heart of the clip.
<<<CHAR_B>>> is generous and sure, like an older sister who already knows the secret: warm, direct, a hint of mischief, never salesy. "Follow me" is half invitation, half literal — she walks off after her friend.
Both keep faces alive between words: micro-reactions, breaths, blinks, moving brows. Nothing posed or frozen.

=== SHOT (15s, one continuous handheld take, no cuts) ===
0.0–0.3s — FRAME 0
Frame 0.0 is exactly <<<image_1>>>, no settle-in. <<<CHAR_A>>> seated low, leaning slightly forward, eyes on the lens. <<<CHAR_B>>> on deck holding the rigging, hair moving in the wind whenever on screen. Camera: handheld low at seated chest height inside the cabin, ~1.2 m from her, never steady.
Sound: water lapping the hull, boat creaking on its lines, halyards ticking, faint wooden cabin echo.
0.3–3.2s — THE HOOK, seated
Leaning in like telling a secret, eyes on the lens, quiet silky American voice: "Everyone's using Opus 5.5 to write emails and code..." ("Opus five-point-five"). Lightly amused, brows lift on EVERYONE'S. On "emails and code" her right hand gives one small dismissive flick by her knee. The "..." is a real half-second beat: lips parted, small inhale, eyes narrowing with a private smile, head tipping a fraction — the good part is coming.
On deck <<<CHAR_B>>> turns her head right, as if called from the quay.
3.2–6.0s — Rising toward the steps
EFFECT: late handheld pan up-left + frame-edge clip
On "and" she rises in one motion, turning to the companionway, looking back over her right shoulder: "...and almost nobody's using it to direct shots like this." Stress on NOBODY with a small conspiratorial chin lift. On SHOTS LIKE THIS her right hand draws one loose circle taking in the cabin, porthole and bright hatch. Her eyes brighten, a real smile starts. Her left hand finds the brass rail.
The camera follows a beat late; she clips the left edge, half her face cut off.
<<<CHAR_B>>> lets go of the rigging and walks off right out of the hatch view. By 6.0s the hatch shows only rigging, sail and bright harbour; two or three gulls far off, small and out of focus.
Sound: seat creak, palm on brass, bare steps fading on teak above, wind through the hatch.
6.0–6.6s — Second step — FOCUS HUNT 1
She steps onto the second step and half-turns back, hatch sunlight across her face and shoulder. The filmer steps closer; as she lands back in frame the autofocus hunts and locks (~1s), silent.
6.6–7.8s — "The skin," — PHYSICAL CLOSE-UP
Softly, almost tenderly: "The skin," — the filmer leans the phone in by hand until the lens is ~25 cm from her face. For about a second the frame is her cheek, nose and eye corner in warm hatch sun: dense freckles, visible pores, fine peach fuzz glowing in backlight, faint flush, tiny unevenness, a stray hair across her cheekbone, the gold chain glinting at the bottom. Focus breathes. Her right fingertips brush her cheekbone, palm toward her face. Her eyes soften and hold the lens with a small private smile. The frame keeps shaking.
This is a hand movement toward her face — never a zoom, crop or punch-in.
Sound: voice suddenly intimate and breathy, hair brushing the mic.
7.8–8.8s — "the light," — PULL BACK + TILT
The phone pulls back to ~1 m and tilts up toward the hatch: low golden sun, glowing white sail, sky over the harbour. Her right hand opens outward toward the hatch, palm up, travelling away from her body. Auto-exposure meters: blown highlights recover detail in sail and sky, cabin sits a touch deeper — a camera metering, never a light change — and rolls back as the frame returns to her. She stays at least partly in frame.
8.8–9.4s — "the grain —"
Back on her at ~1 m: she rubs her right thumb across her fingertips once, like feeling fine sand. Her smile widens.
9.4–10.4s — "even the way the camera shakes." — THE REAL LAUGH
Halfway through she starts laughing, caught off guard; the last words come through the laugh. Behind the camera the filmer lets out one soft wordless breath of a laugh and the frame jiggles and bounces with it for about a second, then returns to normal unsteadiness. Her right hand gives one loose open-handed lift toward the lens and drops — no pointing. Eyes crinkle, head tips back, hair falls across her cheek and she leaves it. Pure unguarded joy — her emotional peak.
10.4–11.6s — OUT THROUGH THE HATCH
EFFECT: focus hunt 2 + auto-exposure cabin → daylight + subtle flare + wind
Still smiling, she climbs the last two steps, left hand sliding up the rail then gripping the hatch coaming, and steps onto the teak deck, moving off left toward the bow without stopping. Autofocus hunts and locks (~1s), silent — her last focus hunt.
Auto-exposure adapts: the white glare of the hatch recovers into real sky, sail, harbour and stone; her skin and linen settle from hot overexposure to correct daylight tone; the cabin below falls into shadow. A slow, silent, gradual roll — never a light change.
A subtle sun flare crosses the lens from the low sun: soft milky wash, faint corner bloom, passing as the frame settles. The instant she is on deck, the sea wind streams her hair sideways.
Sound: strong wind on the mic, fast-clinking halyards, gulls, water slapping the hull, bare feet on teak.
11.6–12.0s — CAMERA ON DECK, TURN RIGHT
The filmer climbs to the top of the hatch. <<<CHAR_A>>> keeps moving left. Without a cut the camera turns slightly right and nearly straightens. Plain hand movement, no sound.
12.0–13.9s — <<<CHAR_B>>> CLOSE — "And I'll teach you how to write it."
She is already in frame the instant the camera lands, standing beside the hatch, having come back from the right side of the deck. FRAMING: ~1 m, head to waist, cut at the top of her white trousers; legs never in frame. Behind her: white sail, harbour wall with lighthouse, low sun, hazy and milky. No dead air. Autofocus locks in ~0.5s, silent.
Looking straight into the lens with warm grey-green eyes, slight Australian accent, easy and certain: "And I'll teach you how to write it." Weight on I'LL and TEACH, a small tip of her head toward herself, a soft knowing half-smile, one small loose hand gesture at waist height. Hair and polo collar whip in the wind; natural blinks.
13.9–15.0s — "Follow me." — HARD OUT
On "Follow me." she is already turning to walk left after <<<CHAR_A>>>, throwing one look back over her shoulder into the lens with a quick wink and a "come on" smile, hair across her face. As she takes her first step away the camera lurches one step forward to follow — cut at that instant, mid-motion. No hold, no fade, no black.

=== ALLOWED CAMERA / LENS BEHAVIOUR (nothing else) ===
Continuous unsteady handheld 0–15s. Late pans (2x). Frame-edge clip (1x). Physical lean-in to her skin (1x). Pull-back + small tilt to hatch (1x). Laugh shake (1x). Auto-exposure adaptation (2x: tilt metering; cabin → daylight). Autofocus hunt on <<<CHAR_A>>> (2x) plus focus breathing in the close-up; on <<<CHAR_B>>> (1x, ~0.5s). Subtle sun flare (1x). Climb to deck + turn right (1x). One forward step at the end. Hard cut on motion at 15.0s (the only cut).
NEVER: other cuts, whips, smears, slow motion, speed ramps, digital zoom, punch-ins, crops, stabilisation, colour grading, added or changing light, transitions, whooshes, shutter or click sounds, music, on-screen text, captions, palm-over-lens, camera turning around, unscripted lines.

=== HARD RULES ===
Exactly 15 seconds, one continuous real-time take. Frame 0.0 = <<<image_1>>> exactly.
Identity lock: <<<CHAR_A>>> always matches <<<image_2>>>, <<<CHAR_B>>> always matches <<<image_3>>>; faces, hair, clothes and jewellery never change. Environment always matches <<<image_4>>>.
SKIN IS NEVER SMOOTH — visible pores, freckles, peach fuzz, fine texture, small unevenness, natural sheen, above all in the "skin" close-up. No smoothing, retouching, waxy or plastic skin, or beauty filter. The single most important rule.
Strong, coarse, visible film grain over the whole frame at all times, heavier in the cabin, still clear on deck. Slight softness and mild compression — real old-phone footage.
Obvious handheld: one hand, constant drift, dips and late corrections, horizon never level for long, never still.
Every movement has a visible beginning, middle and end. Nobody teleports: she rises by rising, climbs by climbing, exits the hatch by climbing through it.
★ NO LIGHT EVER CHANGES. ★ The sun stays low, golden and constant. Every brightness shift is the phone's auto-exposure, slow, gradual and silent. No flash, no light pop, no white frame.
<<<CHAR_A>>>'s gestures are all different and none points at her own chest or torso: flick ("emails and code"), circle ("shots like this"), fingertips on cheekbone ("the skin"), open palm to hatch ("the light"), thumb on fingertips ("the grain"), open-handed lift to lens ("the camera shakes").
<<<CHAR_B>>> exits right at 3.2–6.0s, is off screen ~6.0–12.0s, returns from the right side of the deck framed close, never looks at the lens before her line.
Wind: <<<CHAR_A>>>'s hair still in the cabin, lifting on the steps, streaming hard on deck. <<<CHAR_B>>>'s hair always moving. Gulls audible from the steps on, visible far off, out of focus.
Look: warm low sun raking through the hatch, glossy highlights on varnish and brass, hot golden glow on skin and linen, dark cabin, hazy milky harbour and sky. Candid, amateur, unpolished. No grade, no cinematic look, no text, no music.
Nobody performs for the camera: eyes land on the lens and drift off, except the hook (0.3–3.2s) and <<<CHAR_B>>>'s line, where they hold it warmly.

DIALOGUE — spoken exactly once, exactly as written, in order:
<<<CHAR_A>>>: "Everyone's using Opus 5.5 to write emails and code... and almost nobody's using it to direct shots like this."
<<<CHAR_A>>>: "The skin, the light, the grain — even the way the camera shakes."
<<<CHAR_B>>>: "And I'll teach you how to write it. Follow me."
No added words, repeats or extra lines.

=== AUDIO ===
Raw phone mic only: thin, slightly boxy, light auto-gain pumping, handling noise, wind hitting the mic on deck, no denoising. Camera moves, focus and exposure changes are silent.
<<<CHAR_A>>>: close but roomy in the wooden cabin with a faint short echo; intimate and breathy in the close-up; a little farther on "the light"; real unpolished laugh; thinner and more open on deck.
<<<CHAR_B>>>: same phone mic, same modest level, wind gusting across it.
Palette: hull water, mooring creaks, seat and step creaks, palm on brass, halyards, wind building to strong on deck, gulls, bare feet on teak, hair brushing the mic in the close-up, and faint, easy to miss, a small boat engine idling across the harbour.

=== VOICE ===
<<<CHAR_A>>>: quiet, silky American, warm, talking to one person she likes. Natural speed with small breaths, light vocal fry at phrase ends, a soft mouth click, the real half-second beat after "code...". Line 1 moves from amused to a hushed thrill; line 2 is tender and sensory, collapsing into a genuine laugh.
<<<CHAR_B>>>: slight Australian, warm, grounded, generous — "And I'll teach you how to write it" with quiet certainty; "Follow me" light and playful, half-turned away, carried by the look back and the wink.

AI 실사 영상을 완성하는 3가지 황금 규칙

@abxxai가 정리한 사실적인 비디오 생성의 3대 핵심 원칙은 다음과 같습니다.

1. 결함을 요청하라 (Ask for flaws): 모공, 거친 필름 그레인, 흔들리는 손 떨림, 초점이 흔들리는 포커스 헌팅을 프롬프트에 직접 지정해야 합니다.
2. 조명은 절대 변하지 않는다 (The light never changes): 태양광이나 광원은 고정되어 있으며, 화면의 밝기 변화는 오직 스마트폰 카메라 센서의 '자동 노출 적응(Auto-exposure)'으로만 처리되어야 합니다.
3. 단순한 대사가 아닌 감정 상태를 지시하라 (Give every line an emotion, not just words): 대사마다 눈빛의 떨림, 마이크로 반응, 웃음 터짐, 호흡 타이밍을 세밀하게 지정할 때 비로소 가짜 연기 느낌이 사라집니다.

"말끔한 프롬프트(Clean prompts)가 가짜 영상(Fake video)을 만든다"는 교훈처럼, 물리적 세계의 불완전성과 카메라 장비의 광학적 한계를 언어로 세밀하게 제어하는 것이 초실사 AI 영상을 만드는 핵심 열쇠입니다.

원문 출처

• Abdul Shakoor (@abxxai) X 포스트: everyone's using Opus 5.5 to write emails and code.. almost nobody's using it to shoot real footage