The two-selves pipeline

ONLY YOU CAN SAVE YOURSELF

A man hangs off a sea cliff. The calm version of himself reaches down and pulls him up. One face, two states, fifteen seconds.
documented by SARTHAK
three reference images, one prompt, one generation
Director's Cut
Hanging from the cliff edge, seen from above
The opening frame: the drowning self, seen from where the saved self is standing.

00The idea

A man hangs from a storm-lashed cliff. The man who reaches down and saves him is himself, dry, immaculate, calm. One face, two psychological states, fifteen seconds, and a line that carries the whole idea: only you can save yourself. The film is built from three reference images and one long multi-shot prompt. Every step is below in the order it happened, with the exact prompts.

01The Tools

Seedream
Seedream 5.0 ProEvery still: the location, both character sheets, and the two locked keyframes. One face reference in, 2K out.
Seedance 2.5 on MagnificThe video. Reference mode with three images and one multi-shot prompt, 9:16, 15 seconds, native audio.
Claude
Claude CodeModel: Fable 5. The director: planning the shots beat by beat against the music, writing every prompt, and judging every frame.
ffmpegPulling frames out of every generation at two frames a second to judge them, and the final trim. Free, local.

02Plan the beats before touching a prompt

Fifteen seconds, eight shots, planned against the music before a single prompt was written. The track has exactly two peaks: one at 3.5 seconds, where a wave explodes over the ledge and whites out the frame, and one at 11.5 seconds, where he is hauled up. Everything else is restraint. The grammar: despair is dark, wet, teal and rust; salvation is white, dry and bright; the saviour never hurries; the identity is withheld, hand first, face second, the "it was me" only in the last two seconds. Long shots for the struggle, quick inserts for the hands.

  1. 0 to 3 s, top-down over the ledge, fingers slipping. Tension build.
  2. 3 to 4.5 s, the wave whites out the frame on the loudest hit of the track.
  3. 4.5 to 7 s, spray clears, weaker grip, eyes go up.
  4. 7 to 8.5 s, eye-level close-up, a hand in a white sleeve enters from top right.
  5. 8.5 to 10 s, reverse on the saviour, backlit, hand offered to lens.
  6. 10 to 11.5 s, macro on the two hands locking.
  7. 11.5 to 13 s, the pull-up, shot from below, on the second peak.
  8. 13 to 15 s, the embrace, then the direct look to camera.

03The location

One location for the whole film: a sea arch with a readable upper platform and a lower ledge, so a body can plausibly hang from it. The first take gave me the arch and the storm but built the platform as a concrete slab bolted to the rock. One added sentence fixed it: everything is natural weathered basalt, no straight edges, no man-made structures. The second take is the one every shot was generated against.

Generation 1 of 5
The sea arch location
OUTPUT: the only location. Black basalt arch, natural platform and ledge, steel-blue sea, storm sky.
The location · Seedream 5.0 Pro, 16:9, no references
Create an empty ultra-realistic horizontal 16:9 feature-film location: a colossal wet black stone sea arch above a violent steel-blue ocean, with a readable upper platform of natural rough black rock and a natural lower climbing ledge of jagged wet stone directly beneath it, all geography formed by the sea, entirely natural weathered basalt with no straight edges, no man-made structures, no concrete, no walls, no pier. Enormous waves exploding through the arch, powerful wind, dense sea spray and microscopic droplets, cold silver-gray storm sky. ARRI ALEXA 35 and Cooke 35mm image character, cinematic blue-charcoal grade, organic film grain, 4K. No people, buildings, text, logos, crew, camera gear or backstage elements.

04The drowning self

A character sheet in the vulnerable state: soaked, exhausted, in storm gear: a rust hooded field jacket, a taupe henley, olive cargos, worn boots. The single reference was my locked frontal master from the character sheet page. The prompt never describes the face, the master carries it. Grey background, three views, glasses on. First take, no retries.

Generation 2 of 5
Storm-soaked character sheet
OUTPUT: the drowning self. Wet rust field jacket, taupe henley, olive cargos, worn boots, water on the lenses.
The drowning self · Seedream 5.0 Pro, 2K, ref: frontal master only
Ultra-photorealistic horizontal 16:9 character sheet of the exact same man from the reference image, in a vulnerable storm-soaked state. The reference image is the only identity source: keep his face, glasses, beard, skin tone, age 24 and facial proportions exactly as shown, identical in every view.

Three views on a seamless mid-grey studio background: large wet chest-up portrait on the left, full-body front view standing upright facing camera in the center with head to boots fully visible, full-body back view on the right.

Storm-soaked state: thick black hair soaked, flattened and dripping onto the forehead; exhausted watery eyes behind his thick black square acetate glasses, the lenses beaded with water droplets; water running down the face and beard. Wardrobe: a wet rust-brown hooded canvas field jacket, soaked through and darkened, hood down; a taupe ribbed cotton henley beneath; soaked dark-olive cargo trousers heavy with water; worn dark-brown leather lace-up boots. A small silver stud earring in each ear and a thin black-bead and gold chain at the collar.

Tall man, 5 feet 11 inches, about 88 kg, solid healthy build with broad shoulders and a full chest, not slim, posture heavy with exhaustion, shoulders slightly dropped.

Real water beads on skin, glasses and fabric, wet eyelashes, saturated canvas with heavy wet folds, fine natural skin pores, matte natural complexion, natural anatomy and hands, no beauty filter, no digital smoothing. Unretouched 35mm documentary photography realism, soft diffused studio light.

Finished studio image only, single person shown in three views, no text, no logos, no watermark, no scenery, no props, no production equipment.

05The saved self

The saved self had to read as white, calm and supportive without turning into fashion. The outfit became a long ivory wool overcoat worn open, an ivory crew knit, ivory pleated trousers and suede loafers. The coat is not a style choice. The last beat of the film is the saved self wrapping the wet one, and a long open coat is the garment that does that. Same master, same grey background. First take.

Generation 3 of 5
Ivory coat character sheet
OUTPUT: the saved self. Dry, quiff, open ivory coat, one unbroken block of white at the neckline.
The saved self · Seedream 5.0 Pro, 2K, ref: frontal master only
Ultra-photorealistic horizontal 16:9 character sheet of the exact same man from the reference image, in a calm successful state. The reference image is the only identity source: keep his face, glasses, beard, skin tone, age 24 and facial proportions exactly as shown, identical in every view.

Three views on a seamless mid-grey studio background: large chest-up portrait on the left, full-body front view standing upright facing camera in the center with head to shoes fully visible, full-body back view on the right.

Successful state: completely dry and immaculate. Thick black hair in his voluminous textured quiff swept up and back with short faded sides. Thick black square acetate glasses, clean lenses. Wardrobe: a long ivory wool overcoat, knee length, unstructured with soft natural shoulders, worn open; beneath it a fine ivory merino knit sweater with a plain crew neck; ivory pleated wool trousers with a clean straight drape; ivory suede loafers. A small silver stud earring in each ear. A steel bracelet watch on the left wrist. The neckline is one unbroken block of ivory.

Tall man, 5 feet 11 inches, about 88 kg, solid healthy build with broad shoulders and a full chest, upright unhurried posture, shoulders relaxed and open, expression serene and steady with a faint knowing calm.

Real fabric texture: soft wool nap on the coat, fine knit ribbing on the sweater, matte suede, fine natural skin pores, matte natural complexion, natural anatomy and hands, no beauty filter, no digital smoothing. Unretouched 35mm documentary photography realism, soft diffused studio light with a gentle rim light on the coat shoulders.

Finished studio image only, single person shown in three views, no text, no logos, no watermark, no scenery, no props, no production equipment.

06One face per sheet

A three-view sheet has three faces, and the two small ones on the standing figures are never as accurate as the big portrait. Fed to a video model, the small faces fight the good one. So before anything moved, both sheets were rebuilt: the large portrait kept, the front and back body views cut at the shoulders. The outfit survives, the weak faces do not. This is a crop, not a generation.

Storm sheet with faceless body views
The rebuilt storm sheet. One face, two faceless wardrobe views. The ivory sheet got the same treatment.

07Locking the opening frame

Before the video, I built the opening frame as a still so the camera, physics and face could be judged with no motion to hide behind. Two rules made it work. First, describe the geometry as facts a photographer would state: camera straight down, ledge at the bottom of frame, hands hooked on the edge, body hanging below over the water, one boot kicked out. Second, describe muscles, not feelings. Emotion words get acted out like a cartoon. Brows drawn down and inward, eyes narrowed to slits against the spray, lips parted a finger's width, jaw set, cheeks slack: physical strain with water in the face. The desperation comes from the situation, not from the face performing it.

Generation 4 of 5
Locked opening frame
LOCKED: the opening frame. Camera straight down, hands on the edge, body hanging, face described by its muscles.
The opening frame · Seedream 5.0 Pro, 2K, ref: frontal master
Ultra-photorealistic vertical 9:16 film still shot from directly above a wet black basalt cliff edge, camera looking straight down. The exact same man from the reference image hangs below the edge: both hands hooked over the top of the wet rock ledge at the bottom of the frame, fingers splayed and whitened with pressure, arms extended up toward the camera, head and shoulders just above the edge and tilted up toward the lens, torso and legs hanging below over a violent steel-blue ocean churning with white foam far beneath him, one boot kicked out away from the rock. Keep his face, glasses, beard, skin tone, age 24 and proportions exactly as the reference image. He is soaked: thick black hair flattened and dripping, water beads on his thick black square acetate glasses, water streaming down his face and beard. Wardrobe: wet rust-brown hooded canvas field jacket soaked dark, taupe henley at the collar and cuffs, dark-olive cargo trousers, worn dark-brown boots. Face: head tilted up toward the lens; eyebrows drawn down and inward with a vertical crease between them; eyes narrowed to slits against the spray, lids heavy, water on the lenses; jaw set, lips parted about a finger's width, corners of the mouth neutral, no smile, no grimace, no visible teeth, no crying; cheeks slack with exhaustion. Cold silver-gray storm light, blue-charcoal grade, fine film grain, ARRI ALEXA 35 and Cooke 35mm character, unretouched documentary realism, real water droplets, no beauty filter. Single person, no text, no logos, no watermark.

08Locking the close-up

The shot where the hand appears. Its camera is the saved self's eyes, above at forty-five degrees. The blurred ledge fills the bottom of frame, the hands grip its top edge with fingertips toward the lens, the arms reach forward from the shoulders in the jacket sleeves, the head is nearest, the sea is behind and below. Same method as the opening frame: geometry stated as facts, face stated as muscles, the master carries the identity.

One more decision before it was locked: the shot starts with him looking down at the rock and he only looks up when the hand arrives. A locked first frame decides where the eyes begin, so the still has the eyes down, giving the shot somewhere to go.

Generation 5 of 5
Locked close-up frame
LOCKED: the close-up. Eyes down, ready to look up when the hand arrives.
The close-up frame · Seedream 5.0 Pro, 2K, ref: frontal master
Ultra-photorealistic vertical 9:16 film still, a tight frontal 50mm close-up from the point of view of a person standing on a wet black rock ledge looking down at about forty-five degrees. The exact same man from the reference image hangs from the edge below: the blurred wet rock ledge runs across the bottom of the frame, his two hands grip its top edge with fingertips toward the lens, arms stretched forward from his shoulders in wet jacket sleeves with taupe cuffs showing, his head nearest to camera and tack sharp, his body falling away below, the churning white ocean soft behind and beneath him filling the upper frame. Keep his face, thick black square acetate glasses, short full dark beard, medium-warm brown skin, age 24 and proportions exactly as the reference image. He is soaked: black hair flattened and stuck to his forehead, droplets hanging from his eyelashes and beading on the lenses, water streaming down his face and beard, the collar of a wet rust-brown hooded field jacket over a taupe henley. Face: head bowed a few degrees, eyes cast down at the rock directly in front of him, lids heavy and half lowered, not looking at the camera; brows drawn down and inward, eyes narrowed and wet, lips parted a finger's width, jaw set, no smile, no visible teeth, no crying, exhausted. Cold steel-blue grade, fine film grain, unretouched, no beauty filter. Single person, no second hand in frame yet, no text, no caption, no logos, no watermark.

That is every still this film needed: one location, two sheets, two locked frames. Everything from here is one video generation.

09The video prompt

The prompt is nine timed shots inside the section skeleton I use on every film: scene context, active references with each image's role spelled out, a character block, a direction lock, continuity rules, per-shot timecodes, performance, physics, optics, sound, and a final block of positive locks. Every lesson from the stills is written in: struggle shots top-down with the ocean below and never rolled, he hangs and never climbs, the slip is a full body-width swing, eyes open and light panting, the hand stops clear of the face, the rescue shot from below with bodies filling the frame, wind visibly moving the coat and hair, and the end held on a direct look to camera with a small closed-mouth smile, no fade, because fades belong in the edit. Three images attached in order: the ivory sheet, the storm sheet, the location.

The film · Seedance 2.5 on Magnific, 9:16, 15s, refs: @img1 ivory sheet, @img2 storm sheet, @img3 location
SCENE CONTEXT
15-second ultra-photorealistic multi-shot vertical sequence: a vulnerable young man struggles on a storm-lashed black sea cliff until the calm successful version of himself reaches down and saves him. Prestige psychological feature-film footage.

ACTIVE REFERENCES
@img1 - SUCCESSFUL STATE: left panel is the exact face of the man; the center and right panels are faceless wardrobe views, front and back, of the same outfit. Dry and immaculate: thick black square acetate glasses, short full dark beard, thick black hair in a swept-back quiff, medium-warm brown skin, dark brown eyes, age 24, solid broad-shouldered build; long ivory wool overcoat worn open, ivory crew-neck knit, ivory pleated trousers, ivory suede loafers, steel watch on the left wrist.
@img2 - VULNERABLE STATE of the exact same man: left panel is his face, soaked; center and right panels are faceless wardrobe views, front and back. Same face and glasses, hair soaked and flattened, water beads on the lenses; wet rust-brown hooded field jacket, taupe henley, soaked dark-olive cargo trousers, worn dark-brown lace-up boots.
@img3 - THE ONLY LOCATION: colossal wet black basalt sea cliff and arch, rough natural upper platform, jagged lower ledge, steel-blue ocean, enormous waves, cold silver-gray storm sky.

CHARACTER
Both men are two psychological states of THE EXACT SAME PERSON from @img1 and @img2: identical face throughout, full soft cheeks, rounded jaw under the short full dark beard and moustache, thick black square acetate glasses that never come off either man, small silver stud earrings. Successful man always dry in ivory; vulnerable man always soaked in rust. Never mix outfits. Faces never change shape.

DIRECTION LOCK
In every struggle shot gravity points to the bottom of frame: ocean below the vulnerable man, platform edge above him, head toward the top of frame, boots toward the sea. The camera looks down on him from above the edge, never from the side, never rolled. He hangs from the edge; he never stands on or climbs the cliff face like a wall. The successful man's RIGHT hand is the offered hand; the vulnerable man reaches with his LEFT; his RIGHT stays on the rock. Never reverse hands.

CONTINUITY
Every action shows anticipation, execution, follow-through and a stable ending; cut only after it completes. Preserve body position, hands, screen direction and geography. No jumps, teleportation or frozen gestures.

SHOT 1 - 0.0-1.7 s - DANGER  @img2 @img3
Directly above the wet platform edge. The vulnerable man hangs beneath it, both hands gripping rock, ocean churning far below. He pulls face and shoulders above the edge; one boot loses the rock and kicks out wide, his hips swing a body-width sideways, fingers slide on the wet stone, he drops a few centimeters, then both hands bite in again and the swing dies. Water streams over face, glasses and jacket. 35mm top-down, restrained handheld, immediate vertigo.

SHOT 2 - 1.7-3.2 s - SECOND ATTEMPT  @img2 @img3
From the stable grip he breathes once, hard. Right hand stays planted. Left hand releases, reaches farther, finds a hold, closes. He pulls until his left elbow nearly reaches the platform, shoulders shaking, chest still below the edge, panting. Complete the pull before cutting.

SHOT 3 - 3.2-4.8 s - THE WAVE  @img2 @img3
At his highest point a huge wave explodes up over the edge from behind him and whites out the frame. Spray tears his left hand off the rock; the right hand holds. He swings against the cliff, eyes squeezed shut for the instant of impact, the left hand slaps back onto the ledge, both hands grip, he blinks the water away and pants with his mouth open as the spray clears. He never falls.

SHOT 4 - 4.8-6.3 s - THE UNKNOWN HAND  @img1 @img2 @img3
The successful man's point of view: camera above the vulnerable man at forty-five degrees, tight 50mm close-up. Blurred rock ledge across the bottom of frame, his hands gripping its top edge with fingertips toward the lens, arms forward in the wet sleeves, head nearest and sharp, eyes down on the rock, water streaming down the lenses, blinking. The successful man's RIGHT hand enters cleanly from the upper right, ivory wool coat cuff and ivory knit beneath it, brown skin, open palm offered downward, and stops above and to the right of his head without covering his face. Focus racks to the hand. He registers the movement, lifts his head, eyes rise to the hand and lock on. Brows ease, mouth parts: panic to recognition. Hold.

SHOT 5 - 6.3-7.8 s - SUBJECTIVE REVEAL  @img1 @img3
Reverse: from the vulnerable man's eye line, low 35mm looking up. The successful man crouches at the edge above, exact face, glasses, dry swept-back hair, bright overcast sky behind him. Real wind is on him: the open ivory coat lifts and settles, the lapels flutter, the hem and one loose sleeve billow and drop back, the ribbed knit stays still against his chest; individual strands of hair lift off the quiff and fall back; fine sea spray drifts across the frame between them; his eyes narrow a fraction against it. He looks down at his vulnerable self with calm certainty, steady gaze, the faintest reassuring nod. His right arm completes the extension and waits, the coat cuff moving on his wrist. No facial distortion.

SHOT 6 - 7.8-9.2 s - THE GRIP  @img1 @img2
Macro on exactly two active hands. The successful RIGHT hand, ivory wool cuff, offered; the vulnerable LEFT hand, wet rust sleeve, rises to it while his right stays on the cliff. Palms approach, fingertips touch, wet fingers slide, then successful right hand closes around vulnerable left wrist and vulnerable left hand closes around successful right wrist. One clear right-to-left wrist grip, never reversed. Five correct fingers per hand, natural thumbs, tendons, skin compression. Hold the grip before pulling.

SHOT 7 - 9.2-11.8 s - COMPLETE RESCUE  @img1 @img2 @img3
From below the ledge looking up, bodies filling the frame. The successful man braces both feet, lowers his center and pulls, left hand on rock. The vulnerable man keeps his right hand on the edge and pushes with both boots; one slips, slides, finds a crack. His shoulders rise above the platform; the successful man hooks his free left arm under the vulnerable upper arm and pulls his chest over the edge. One knee, the other, both boots on the platform. Weight and effort in both faces. No cut until his whole body is safe. No distant side view.

SHOT 8 - 11.8-13.3 s - COMPLETE EMBRACE  @img1 @img2 @img3
The vulnerable man's knees soften. The successful man catches him, brings him upright, one arm around his shoulders, the other hand behind the wet head. The open ivory coat closes around the soaked rust jacket. The vulnerable man wraps both arms around him, rests his head on the shoulder, closes his eyes and keeps them closed, face releasing into exhausted relief. Bodies settle; water transfers onto the ivory wool. Protective, brotherly, never romantic or posed.

SHOT 9 - 13.3-15.0 s - THE LOOK  @img1 @img2
Same framing as the embrace, camera does not move, no push-in. The two men stay locked together, the vulnerable man's face buried in the ivory shoulder with eyes closed. The successful man holds the hug for a beat, then turns only his head toward the lens and gives a direct, unblinking, dead-steady look straight into the camera. Then a very small smile: lips closed, no teeth, the corners of the mouth lifting a few millimeters, the eyes barely changing, calm and certain. Hold to the last frame. Wind still moves his hair and the coat. The clip ends on the held look, full exposure, no fade, no transition. Never end mid-movement.

PERFORMANCE
Vulnerable man: losing, not climbing; every action a failure caught at the last instant; weak, heavy, shaking; eyes open and narrowed against the spray except the wave instant; brows down and inward; lips parted, panting lightly; no crying, no bared teeth. Successful man: slow, calm, certain, dry; micro-expressions not gestures; steady eye contact; wind visibly moves his coat and hair in every shot; final smile closed-mouth, no teeth, a few millimeters at the corners, eyes steady.

PHYSICS
Real time, no slow motion. Real weight in every grip, slip and pull: shoulders load, fingers whiten, wet canvas clings, water streams and drips at real speed, the wave arrives as a solid mass then spray then runoff. Wool coat swings with wind and body.

OPTICS AND IMAGE CHARACTER
ARRI ALEXA 35 LogC4 character; Cooke 35mm for action and subjective angles, 50mm for faces and hands; T2.0-T2.8, soft highlight roll-off, dimensional skin, gentle halation, fine film grain, controlled handheld weight. Steel-blue ocean, blue-charcoal shadows, silver-gray sky, ivory wool, earthy rust and olive, natural warm brown skin, Kodak 2383 print character, readable blacks, no plastic sharpness.

SOUND
Realistic diegetic audio only, no music: wind, deep ocean roar, the wave impact at 3.2 s, wet hands on stone, boots slipping, soaked fabric, breathing and light panting; close breathing and dripping water during the embrace. No dialogue, no voice-over.

TECH
Photoreal, unretouched, real physics and motion blur, vertical 9:16, 15 seconds, 24 fps, real time.

POSITIVE LOCKS
Both men are the man from @img1 and @img2, glasses on every frame, no face drift. Ivory outfit only on the dry man, rust outfit only on the wet man, one location from @img3. Struggle shots top-down with the ocean below. Right hand from above meets left hand from below, one clear grip. Every reach, slip, grip, pull and embrace completes before the cut. Final frame: completed embrace, no push-in, direct look to lens, small closed-mouth smile, held at full exposure to the last frame, no fade to black, no transition effects anywhere in the clip. Only the story world: no camera, rig, crew or lights. No extra people or hands, five correct fingers per hand, no text, no captions, no logos, no watermark.

10What not to do

  1. Do not describe emotions. "Terror" becomes a grin and "on the verge of tears" becomes bawling. Describe the muscles: brows, lids, lips, jaw.
  2. Do not describe a composition with adjectives. State it as a photographer would: where the camera is, what fills the bottom of frame, where the hands are, what is behind the head.
  3. Do not attach three faces when you mean one. Cut the small faces off the body views and keep one strong portrait per sheet.
  4. Do not lock a first frame at the end of an action. If the shot is a look up, the frame must start with the eyes down.
  5. Do not add beats the film does not need. A cough, a stumble, a gesture: every extra action is another thing the model can get wrong.
  6. Do not say "wind" or "cinematic". Say what moves: lapels flutter, hem lifts and drops, strands of hair leave the quiff and fall back.
  7. Do not put fades or transitions in the prompt. End on the held frame and do the fade in the edit where you control it.

One face. Two states. One cut.

The plan did all the directing. My job was to write it down precisely and then get out of the way.

The ride into war Read next The Ride Into War A present day ride that match-cuts into a Rajput battle. → The character sheet Read next The Character Sheet How to make AI hold your face. Every view, every prompt. →