AI로 스파이더맨처럼 날아다니는 영상 만들기

요즘 스파이더맨 영화가 너무 유행인데요. AI를 활용하면 그런 영상을 만드는 것도 충분히 가능하더라고요. 그 과정을 정리해봤습니다.

1. 레퍼런스 조사

가장 먼저 했던 일은 기존에 잘 만들어진 영상을 찾는 일이었습니다!

저는 해외 계정에서 잘 만든 영상을 찾고, 그 프롬프트를 제공해주는 곳에 댓글을 달아서 프롬프트 정보를 받았어요.

https://www.instagram.com/reels/Db1xpyQyXb_/

위 링크를 보시면 제가 만든 영상과 유사한 움직임의 영상을 제작하고, 프롬프트를 제공해주었길래 이 프롬프트를 받아서 베이스로 활용했습니다.

2. 캐릭터 세팅

그 다음은 저의 캐릭터를 세팅하는 것이죠. 클로드와 함께 제 사진을 넣고 저의 다양한 각도의 얼굴과 의상을 세팅했어요. 아주 마음에 들거나 완전히 똑같지는 않지만, 저의 개성을 어느정도는 담아주어서, 이걸로 진행하기로 결정 했습니다.

3. 소품 세팅

이 영상의 포인트는 처음에 가면을 쓰는 장면이었는데요. 그 가면을 세팅할 필요가 있었죠.

어려운 프롬프트는 아니었어요 ㅎㅎㅎ "아주 멋쟁이 마스크 디자인 해달라고 ㅎㅎㅎ"

근데 사실은 제가 스파이더맨이 되고 싶긴 했었는데, 이게 저작권 때문에 막혔더라고요.

그래서 이 부분은 포기하기로 하고, 멋쟁이 마스크로 만족해야 했네요 ㅎㅎ

4. 영상 제작

위 계정에서 받았던 프롬프트가 있었기에, 그 프롬프트를 베이스에 주고 제 내용에 맞게 수정을 해서 바로 영상을 제작 해봤어요. 한번 보여드릴게요.

사실 처음에 제 얼굴이 너무 비슷하게 나와서 마음에 들었고,

30초 짜리 영상이 한번에 만들어진다는것도 놀랐습니다.

(SEEDANCE 2.5 가 이렇게 강력하더라고요)

하지만, 중간 중간에 보면 회전을 하거나 할때 갑자기 몸이 뭉게 지기도 하고, 움직임이 역동적이지 않고 조금 딱딱한감이 있더라고요. 그래서 중간에 한 단계를 더 짚어 넣었어요.

5. 액션 시트 제작

바로 액션 시트를 만드는 거에요. AI에게 움직임이 있을 때 이걸 참고하라고 해주는 거죠.

그렇게 나온 액션 시트를 만들었고요. 이 액션시트를 기반으로 다시 한번 영상을 만들어봤어요.

이번에는 앞에 부분은 제외하고 창문으로 떨어진 이후의 뒷 부분 20초만 다시 만들어봤어요~!

어떤가요? 움직임이 훨씬 더 자연스럽게 나오지 않았나요?

그래서 최종적으로는 이 2개의 영상을 편집해서 완성을 했고요.

완성본은 제 인스타그램에서 봐주시길 부탁드립니다 ㅎ

>> https://www.instagram.com/p/Db5OSN9OAsO/

그렇게 만들어진 최종 프롬프트 2종을 공유합니다 !

[Prompt 1. 처음에 창가에 앉아서 가면을 쓰는 장면 프롬프트]

CHARACTER REFERENCE - MANDATORY: The man in this film is @image1. His face, hairstyle and facial features must match @image1 exactly in every frame where his face is visible. His body, build and wardrobe must match @image2 exactly: Korean man, reads early-to-mid thirties, slim build, short black hair neatly styled, clean-shaven, calm confident face. Outfit exactly as @image2 - an all-black tailored suit: slim black dress trousers, black dress shirt with no tie and the top button open, fitted black suit jacket, and a LONG BLACK WOOL OVERCOAT reaching below the knees worn open, the coat tails free to whip and stream in the wind, black leather dress shoes. HE and HIM. Male build, male silhouette, male facial structure.

THE MASK is @image3 and must match @image3 exactly in every frame where it is worn: a matte black full-head technical balaclava hood in fine ripstop grid fabric, covering the entire head and neck, no skin, no hair, no mouth opening, with a NARROW WRAPAROUND VISOR in mirrored deep crimson red sweeping continuously from temple to temple like sleek sport eyewear, hugging the head closely, a shaped notch at the nose bridge, NO frame or bezel, its edge meeting the black fabric in a clean seamless join, a single thin crimson seam from the crown down the back of the head, low ribbed collar tucked into his shirt. NO web pattern, NO radial lines, NO spider motif, NO emblem, NO logos, NO blue. An original design. The mask looks IDENTICAL to @image3 in every frame.

CRITICAL - THE MASK IS OFF AT THE START. For the first section his bare face is fully visible, exactly matching @image1. He is holding the mask from @image3 rolled up in his hands, not wearing it. He only puts it on after he speaks. Once it is on, it NEVER comes off for the rest of the film.

CRITICAL - CHARACTER ANATOMY. NON-NEGOTIABLE: Athletic, lean, controlled. NEVER ragdoll. NEVER limp. NEVER floppy. Core always engaged. Every movement has muscle intention. He moves like a trained climber and trail runner who happens to have superpowers - precise, casual, effortless, and immaculately dressed. Body follows physics logically. His long coat follows real cloth physics at all times.

CRITICAL - PHYSICAL CONTACT WITH ENVIRONMENT. NON-NEGOTIABLE: In the airborne section he does NOT just fly through the air. Feet push off walls, hands grab ledges, body grazes surfaces. At every contact point knees absorb, muscles engage, the body reacts. This is parkour plus web-swinging, not floating.

CRITICAL - BODY POSITIONS PER MOVEMENT TYPE. Each is a real athletic position and each has a reference panel in @image4. Match the body in the panel, not just the words:

Web swing and firing a web, MATCH @image4 panel A1 and panel A5: both arms extended straight up and forward overhead gripping the web, shoulders stretched, back straight, core tight, legs together and trailing, toes pointed, chin up, the body forming a clean diagonal under the line, coat streaming behind him like a cape. When firing, one arm thrusts out and a white line launches from the hand exactly as in panel A5.

Wall run, MATCH @image4 panel A2: body leans 45 degrees toward the wall, feet striking in a running gait.

Wall push-off, MATCH @image4 panel A3: explosive two-foot push off the wall edge, body launching away.

Street sprint, MATCH @image4 panel A4: full sprint, deep forward lean, arms driving, head forward, body leaning into travel, coat tails snapping behind him.

Backward fall: back to the ground, chest facing up at the sky, arms loose and slightly out, legs relaxed, committed and unhurried, coat tails streaming upward past his shoulders.

Mid-air recovery: one shoulder drops and drives across the body, hips follow, the torso rotates 180 degrees in one clean controlled twist until he is face-down.

The final rise, MATCH @image4 panel A6: floating upright at the peak, arms open wide, body arched, everything released.

CRITICAL - THE ROOM. A densely cluttered lived-in New York bedroom, late afternoon, overcast. Dead centre, a large black-framed sash window filling most of the frame, with a deep sill he is sitting on. Beyond the window: sky BLOWN OUT to near-white with soft highlight rolloff, no sun anywhere, and low-rise red-brick residential New York rooftops with wooden water towers, fire escapes and a hazy distant skyline. Left: shelves packed tight with paperback spines, a desk buried in stacked books, a wooden stool. Right: a black metal bed frame with rumpled grey and blue bedding, a coat hanger with a spare suit jacket, generic unbranded posters. NO recognisable comic covers, NO movie posters, NO legible logos or title text anywhere.

CRITICAL - CAMERA. For the first section (0-14.5 sec) the camera is completely locked. It does not move, pan, tilt, zoom, push, drift or shake. Perfectly symmetrical composition, the window dead centre of frame, lens at seated eye level. A single static locked-off wide shot. At 14.5 sec, the moment he tips out of the window, the camera UNLOCKS and DIVES FORWARD THROUGH THE OPEN WINDOW after him in one continuous move, converting into an FPV drone-follow - extreme wide-angle, very low altitude, aggressive real motion blur, constant physical roll and bank - for the rest of the film. ONE continuous take, NO cuts anywhere.

CRITICAL - REALISM AND FILM STOCK, AND IT MUST SURVIVE THE WARM GRADE. NON-NEGOTIABLE: Real Hollywood film. RED Monstro 8K, anamorphic. This is 35mm film from the first frame to the last. Visible organic grain sits over EVERY frame, at its most obvious in the warm midtones of the golden-hour section. Heavy warm halation blooms off every bright highlight. Deep filmic highlight rolloff so the sky never clips to flat white. Lifted, slightly desaturated blacks. The bedroom section is cool, flat, desaturated overcast daylight - no direct sun, no hard shadows, NEVER golden hour indoors. As the grade warms toward golden hour in the swing section the image must get MORE filmic, not less: more grain, more halation, more rolloff. NEVER clean. NEVER glossy. NEVER digital. NEVER an over-saturated orange-and-teal sunset. ALL MOTION PLAYS AT NORMAL REAL-TIME SPEED WITH FULL NATURAL GRAVITY. NEVER slow motion, NEVER ramped, NEVER speed-changed at any point.

CRITICAL - THE WEB MUST BE VISIBLE. NON-NEGOTIABLE: A bright white web line is clearly visible in almost every airborne shot, exactly as shown in @image4 panels A1 and A5. It is a real physical rope of white silk, thin and taut and catching the light, running from his hand up to a building edge, anchored at both ends, bearing his full body weight. When he is in the air he is HANGING FROM A WEB, swinging on an arc like a rope swing. He is NEVER in unsupported freefall except immediately after leaving the window. NEVER wingsuit. NEVER skydiving. NEVER belly-to-earth. NEVER a bare parkour jump.

30-second continuous shot. ONE take. NO cuts.

0.0-2.5 sec: He sits fully up on the window sill, dead centre of frame, both feet up off the floor, knees drawn up, ankles crossed, the long coat draped over the sill. He holds the rolled-up black mask from @image3 in both hands in front of his chest, looking down at it. His bare face is clearly visible, matching @image1. Quiet and still. Behind him the blown-out white sky and the brick rooftops.

2.5-4.0 sec: He lifts his head and looks straight down the lens. Calm and settled, a small private smile. His face fully visible, matching @image1.

4.0-6.0 sec: HE SPEAKS, calm and naturally, with a bit of excitement. His lips and jaw clearly and visibly articulate the words: "Okay, let's do this!" Natural, not shouted, not muttered, not slow. His mouth must visibly form each word, accurate lip sync. A small composed smile and a slight nod on the last word. His face is unmasked and fully readable throughout this beat.

6.0-9.0 sec: He raises the mask in both hands and PULLS IT DOWN OVER HIS HEAD in one smooth continuous real-time motion. The fabric stretches over his crown, down past his face, and settles over his neck, the crimson visor coming to face straight down the lens, mirrored and unreadable, now matching @image3 exactly. He smooths it once at the jaw, then straightens his coat lapels with both hands - one sharp tailored tug. CRITICAL: the mask is physically pulled on, the fabric visibly stretching and moving over his head. It must NOT simply appear or cut into place.

9.0-14.5 sec: A steady breath. He tips backward out of the open window in ONE smooth continuous real-time motion. He simply lets his weight go and gravity takes him - hips lead, shoulders follow, his legs swing up and over the sill in a single fluid arc, coat tails the last thing to whip over the edge. Quick, weighted, exactly the way a real body falls backward off a ledge. NEVER slow motion. NEVER a floaty suspended hang. The camera holds one beat on the empty sill - then DIVES through the window after him.

14.5-16.0 sec: FPV now. He is FALLING BACKWARD, back to the ground, chest facing up at the sky, arms loose, coat streaming upward past his shoulders. Directly above him and shrinking fast is the open window he just left. The camera falls with him. Brick facades rush past on both sides. He is calm. He is not flailing.

[Prompt 2. 창문 뒤로 떨어진 후 뉴욕 시내 날라다니는 부분 프롬프트]

CHARACTER REFERENCE - MANDATORY: The figure in this film is the man from @image1, and his body, build and wardrobe must match @image2 exactly: Korean man, slim athletic build, wearing an all-black tailored suit - slim black dress trousers, black dress shirt with no tie, fitted black suit jacket, and a LONG BLACK WOOL OVERCOAT reaching below the knees worn open, the coat tails whipping and streaming in the wind, black leather dress shoes. HE and HIM. Male build, male silhouette. When his face is revealed in the final beat, it matches @image1 exactly.

THE MASK is @image3 and he wears it for the entire film UNTIL THE FINAL BEAT, where he pulls it off: a matte black full-head technical balaclava hood in fine ripstop grid fabric, covering the entire head and neck, no skin, no hair, no mouth opening, with a NARROW WRAPAROUND VISOR in mirrored deep crimson red sweeping continuously from temple to temple, a shaped notch at the nose bridge, seamless join with the black fabric, a single thin crimson seam from the crown down the back of the head, low ribbed collar tucked into his shirt. NO web pattern, NO radial lines, NO spider motif, NO emblem, NO logos, NO blue. The mask must match @image3 exactly in every frame where it is worn.

CRITICAL - CHARACTER ANATOMY. NON-NEGOTIABLE: Athletic, lean, controlled. NEVER ragdoll. NEVER limp. NEVER floppy. Core always engaged. Every movement has muscle intention. He moves like a trained climber and trail runner who happens to have superpowers - precise, casual, effortless, and immaculately dressed. Body follows physics logically. His long coat follows real cloth physics at all times.

CRITICAL - PHYSICAL CONTACT WITH ENVIRONMENT. NON-NEGOTIABLE: He does NOT just fly through the air. Feet push off walls, hands grab ledges, body grazes surfaces. At every contact point knees absorb, muscles engage, the body reacts. This is parkour plus web-swinging, not floating.

CRITICAL - BODY POSITIONS PER MOVEMENT TYPE. Each is a real athletic position and each has a reference panel in @image4. Match the body in the panel, not just the words:

Backward fall: back to the ground, chest facing up at the sky, arms loose and slightly out, legs relaxed, committed and unhurried, coat tails streaming upward past his shoulders

Mid-air recovery: one shoulder drops and drives across the body, hips follow, the torso rotates 180 degrees in one clean controlled twist until he is face-down

Web swing and firing a web, MATCH @image4 panel A1 and panel A5: both arms extended straight up and forward overhead gripping the web, shoulders stretched, back straight, core tight, legs together and trailing, toes pointed, chin up, the body forming a clean diagonal under the line, coat streaming behind him like a cape. When firing, one arm thrusts out and a white line launches from the hand exactly as in panel A5.

Wall run, MATCH @image4 panel A2: body leans 45 degrees toward the wall, feet striking in a running gait

Wall push-off, MATCH @image4 panel A3: explosive two-foot push off the wall edge, body launching away

Street sprint, MATCH @image4 panel A4: full sprint, deep forward lean, arms driving, head forward, body leaning into travel, coat tails snapping behind him

The rise, MATCH @image4 panel A6: floating upright at the peak, arms open wide, body arched, everything released

Rooftop landing: drops the last two metres, lands two-footed on gravel, knees bending deep to absorb, one hand touching down for balance, coat settling around him a half-second later

CRITICAL - REALISM. RED Monstro 8K, anamorphic, flown as an FPV drone - extreme wide-angle, very low altitude, aggressive real motion blur, constant physical roll and bank. ALL MOTION AT NORMAL REAL-TIME SPEED. NEVER slow motion.

CRITICAL - FILM STOCK, AND IT MUST SURVIVE THE WARM GRADE. NON-NEGOTIABLE: This is 35mm film from the first frame to the last. Visible organic grain sits over EVERY frame, and it is at its most obvious in the warm midtones of the golden-hour section, not just in the cool streets. Heavy warm halation blooms off every bright highlight. Deep filmic highlight rolloff so the sky never clips to flat white. Lifted, slightly desaturated blacks. As the grade warms toward golden hour the image must get MORE filmic, not less: more grain, more halation, more rolloff. NEVER clean. NEVER glossy. NEVER digital. NEVER an over-saturated orange-and-teal sunset. The grain must match the opening clip so the two cut together invisibly.

CRITICAL - THE WEB MUST BE VISIBLE. NON-NEGOTIABLE: A bright white web line is clearly visible in almost every airborne shot, exactly as shown in @image4 panels A1 and A5. It is a real physical rope of white silk, thin and taut and catching the light, running from his hand up to a building edge, anchored at both ends, bearing his full body weight. When he is in the air he is HANGING FROM A WEB, swinging on an arc like a rope swing. He is NEVER in unsupported freefall except in the first two seconds and the final two-metre drop onto the rooftop. NEVER wingsuit. NEVER skydiving. NEVER belly-to-earth. NEVER floating. NEVER a bare parkour jump.

20-second continuous FPV drone-follow shot. NO cuts. This is a DIRECT CONTINUATION of the previous shot, in which he tipped backward out of a high window. It opens already mid-fall. It does NOT open on black.

0.0-1.0 sec: He is FALLING BACKWARD, back to the ground, chest facing up at the sky, arms loose, coat tails streaming upward past his shoulders. Directly above him and shrinking fast is the open window he just left, a small dark rectangle with a curtain still lifting in it. The camera falls with him. Brick facades rush past on both sides. He is calm. He is not flailing.

1.0-1.8 sec: THE RECOVERY. One shoulder drops and drives across his body, hips follow, and his torso rotates 180 degrees in one clean controlled twist until he is FACE-DOWN over the street. The camera rolls with him so the city inverts around the frame. One continuous athletic movement, not a flip and not a tumble.

1.8-2.5 sec: HIS ARM PUNCHES FORWARD AND UP, MATCHING @image4 panel A5. Camera whips in tight on the hand, fingers splayed, wrist snapping back. A BRIGHT WHITE WEB FIRES from the hand and shoots up out of frame, the brightest thing in the shot. The line SNAPS TAUT with a hard jolt that visibly yanks his shoulder and arrests his fall, and his drop converts into a forward swinging arc. Camera pulls back fast.

2.5-5.0 sec: WIDE - residential street. Brownstones, fire escapes, awnings, autumn trees. The taut white line runs from his hand up to a rooftop and he SWINGS FROM IT LOW, 3 metres above the traffic, body MATCHING @image4 panel A1, coat streaming behind him like a cape. At the bottom of the arc his black dress shoes skim the roof of a parked car, brief real contact, feet bounce off, he uses it to redirect. Camera banks 40 degrees.

5.0-7.5 sec: He releases at the peak, then RUNS ALONG a brick wall, 4 to 5 quick steps up the vertical surface MATCHING @image4 panel A2, then pushes off hard with both feet MATCHING @image4 panel A3, and A NEW WHITE WEB FIRES from his hand, striking a building across an intersection. The line snaps taut and he swings from it, body matching panel A1. Camera rolls 60 degrees.

7.5-10.5 sec: Still hanging from the visible line he comes in LOW up a narrow street packed with yellow taxis, 2 metres off the ground. The web releases and whips away. His dress shoes hit wet asphalt, full sprint MATCHING @image4 panel A4, 3 to 4 steps between the cabs, coat tails snapping behind him. A horn blares. He laughs. Then he plants one foot on the flank of a taxi, pushes off the roof, and A WHITE WEB FIRES UPWARD, the line going taut and hauling him up as he rockets HIGH. The grade begins to warm.

"WOOHOO!"

CRITICAL - FRAMING FOR THE SWING FINALE. NON-NEGOTIABLE: From 10.5 to 15.5 seconds he is LOCKED DEAD CENTRE OF FRAME. The camera holds him pinned in the middle and never lets him drift off centre. He does not move around the frame - the CITY moves around HIM. Buildings, rooftops and sky sweep and fall away past him while he stays fixed in the centre. He is small and the sky is enormous.

10.5-11.4 sec: His arm punches up MATCHING @image4 panel A5 and A WHITE WEB FIRES, striking high on a building ahead. The line SNAPS TAUT and his run converts into a swing as his feet leave the ground. He is centre frame. The camera pins him there and drops with him.

11.4-12.6 sec: THE BIG SWING. He falls through a long deep arc on the taut line, body MATCHING @image4 panel A1 - arms extended overhead gripping it, legs trailing, core tight - the long black coat carving a clean line through the air behind him. He passes through the bottom at speed and rises up the far side. He stays LOCKED CENTRE FRAME while the whole city sweeps past behind him. Golden hour light rakes across him.

12.6-13.4 sec: HE LETS GO. Both hands open and release the web at the top of the rise. The white line whips away. All of the swing's momentum now carries him UP and FORWARD. His body extends, arms coming down and out. Still dead centre.

13.4-15.5 sec: HE SOARS, body MATCHING @image4 panel A6 - upright at the peak, arms open wide, body arched, everything released. He rises and clears the roofline of a building beneath him, parapet and water tower sweeping past under his feet. The camera cranes up with him, holding him dead centre. Behind him the city opens out - rooftops, water towers, the low sun on the horizon, enormous clouds lit from beneath with orange and pink, thick warm haze. At the very top of the climb his momentum dies to a weightless instant.

15.5-17.0 sec: THE LANDING. Gravity takes him and he drops the last two metres onto the gravel rooftop of the building he just cleared, landing two-footed, knees bending deep to absorb, one hand touching down for balance, coat settling around him a half-second later. Real weight, real impact. The camera sweeps down and around in front of him, decelerating, settling into a steady chest-height medium shot facing him. The wind dies down.

17.0-18.5 sec: He rises to standing, unhurried, golden hour light raking across him, the city skyline glowing behind him. He reaches up with both hands, grips the mask at the jaw, and PULLS IT OFF upward in one smooth real-time motion - the fabric stretching and peeling over his chin, his face, his crown, his black hair springing free slightly tousled. The mask comes away in his hand. His bare face is revealed, matching @image1 exactly. CRITICAL: the mask is physically peeled off, fabric visibly stretching. It must NOT simply vanish or cut away.

18.5-20.0 sec: He lowers the mask to his side in one hand, exhales once - a small satisfied breath, shoulders dropping - and looks STRAIGHT DOWN THE LENS. Face fully visible, matching @image1, hair slightly windblown, a calm confident half-smile. He holds the look. The camera holds perfectly still. Behind him the sun sits low between the water towers. The frame holds. Cut.

[END]

AUDIO: SFX ONLY. NO MUSIC. NO voiceover, NO narration, NO lyrics, NO spoken words other than the single shout. Open on wind roar already at full volume, continuing from the previous shot. Wool coat snapping in the wind. The whip and thud of the web firing and going taut. Web twang under load. Dress shoes on a car roof. Dress shoes on wet asphalt. A taxi horn. His laugh, muffled inside the mask. "WOOHOO!" echoing between buildings. The line singing under load through the big final swing, then a sharp snap as it releases. The wind thins out as he rises, a weightless beat of near-silence at the peak - then a solid two-footed CRUNCH on rooftop gravel, coat fabric settling. Wind drops to a soft rooftop breeze, distant city hum far below. The stretch and rustle of fabric as the mask peels off. One audible exhale. Quiet hold to the cut.