Transform
The Full AI Workflow Breakdown
Everyone in the shot runs from the explosion. One guy walks. He kneels, puts his palm on the pavement, and the machine takes him, then he sprints straight into the smoke everyone else fled. Fifteen seconds, one Seedance 2.0 generation, no editing tricks.
The workflow: Claude writes and directs the prompts, Higgsfield generates every still and the final video, and I cut it in my editor. Steal the prompts below, swap in your own face and your own character, and build your version. That is the whole point of this page.
Transform. One 15 second Seedance 2.0 generation, mastered at 4K. Built inside Higgsfield.
Everything on this page is free. The pack has the exact reference images, the robot sheet, the location plate, and every prompt as a numbered text file, including the character sheet template.
Download the Workflow Pack zip, 54 MB: references, build assets, 10 prompt files including both AI versions, README
Stage 1: The Assets
Every image below was approved as a still before any video credits were spent. Stills cost about 7 credits. Video rolls cost 67 to 330. Lock the pictures first.
Your character sheet
What it does: this is the identity law. It rides along on the final video generation so your face does not drift mid shot.
My character sheet. Multi-view, plain gray background, the wardrobe I actually wear.
CHARACTER SHEET TEMPLATE, make one of yourself
Multi-view character reference sheet of the exact person in the reference photos: identical face, identical hairstyle. Panels: full-body front view and full-body back view standing in a neutral A-pose on the left; large front head-and-shoulders portrait, back of head, and left and right profile views on the right. Wardrobe in every panel: [YOUR SHIRT], [YOUR PANTS], [YOUR SHOES]. Even, soft studio lighting on a plain mid-gray seamless background, photorealistic, crisp detail, consistent scale between panels. No text, no labels, no watermarks, no film borders.
Attach four to six clear photos of yourself and fill in the wardrobe brackets. Keep the background plain gray. That is the whole trick.
The robot character sheet
What it does: defines the machine you become. Front and back view so the generator knows the whole costume.
The robot soldier sheet. Mechanical feet, rifle, katana on the back.
ROBOT CHARACTER SHEET
Two-view character reference sheet of a robot soldier: identical helmet with glowing teal visor, gunmetal-black segmented armor, teal chest plate with a metallic emblem and matching teal back plate, teal shin and forearm accents. Front view on the left, back view on the right, both full-body, standing in the same neutral A-pose, on a dark charcoal studio background with soft even lighting. Fully robotic mechanical feet in the same design language as the articulated mechanical hands: segmented gunmetal plating flowing from the shin armor over the ankle into articulated metal toes, exposed pistons at the ankle, no boots, no fabric, just machine. A sleek futuristic carbine rifle held in the right hand at low-ready in the front view. A high-tech sword with a dark hilt sheathed in a black scabbard mounted diagonally across the back. Photorealistic, crisp product-sheet clarity. Clean full-bleed image: no film borders, no light leaks, no text or labels other than the emblem on the armor.
The emblem on the chest is my logo. Swap in yours or delete that line.
The location plate
What it does: locks the street. Every keyframe references this image so the world stays the same from shot to shot.
The master location plate. Bright daylight, dust wall backlit at the end of the block.
LOCATION PLATE
Master location plate, cinematic film still. A Manhattan side street on a bright, crisp late-morning — strong clean sunlight raking across aged brick facades, fire escapes, glass storefronts throwing sharp reflections, a sliver of vivid blue sky between the rooflines. Parked cars along both curbs, steam wisping from a manhole grate. Far down the street, a huge dust shockwave from an unseen explosion rolls between the buildings toward camera — but it is brightly BACKLIT by the sun, glowing warm cream and gold against the blue sky, beautiful and wrong at the same time. Scattered papers mid-air on the wind. Palette: sunlit warm brick and cream dust, crisp blue sky accent, clean neutral asphalt. Eye-level view straight down the sidewalk. Photorealistic 35mm cinema still, fine film grain in texture only, deep focus. No people anywhere, no vehicles in motion. Full-bleed edge-to-edge image, no film border, no vignette frame, no rounded corners, no light leaks, no text, no watermark.
Keyframe 1: the crowd hook
What it does: the opening frame of the video. The camera hides inside the fleeing crowd and finds you through the gap.
Found in the crowd. This exact image is the START frame of the final generation.
KEYFRAME, CROWD HOOK
Cinematic film still, SHOT 1 of a storyboard — THE HOOK. Setting: exactly the bright Manhattan street from the first reference image — same sunlit brick, blue sky, golden backlit dust wall down the street, parked cars, bare trees. Camera is EMBEDDED INSIDE a fleeing crowd at shoulder height, 135mm long lens, running with the panicked people AWAY from the dust wall: the frame edges are filled with out-of-focus jostling bodies photographed from behind — a blurred shoulder and backpack strap slicing the left edge, a woman's whipping hair and a man's suit shoulder smearing the right edge, heads bobbing, heavy motion blur on every runner. Through a narrow GAP between the blurred bodies, in crisp focus in the middle distance: the man from the second reference image walking calmly in the OPPOSITE direction, toward the golden dust wall, compressed by the long lens, the only sharp and still thing in a frame full of panic. His expression: level, unhurried, eyes forward. Papers flying overhead in the sunlight. 16:9, photorealistic, shallow depth layers, fine film grain in texture only, full-bleed edge-to-edge, no film border, no vignette frame, no light leaks, no text, no watermark.
Keyframe 2: the feet
What it does: pins the ground-level shot where you stop and kneel. The transformation starts from this camera position.
Sneakers walking into the low frame. The stop happens here.
KEYFRAME, FEET
Cinematic film still, SHOT 2 of a storyboard — THE STOP. Setting: exactly the bright Manhattan street from the first reference image — sunlit asphalt, warm golden dust haze drifting low, parked cars, papers skidding in the wind. Extreme low angle, camera lens inches above the asphalt, 35mm, directly ahead of a man's feet walking TOWARD camera, framed shin-down, deep background bokeh of fleeing pedestrians' blurred legs rushing the opposite way and the golden dust wall glowing between the buildings. Sharp focus: the sneakers from the second reference image mid-stride toward camera, jeans hems, natural walking gait, both feet pointing toward camera on the same line of travel — the front foot just PLANTING firmly, weight settling, as if about to stop. Grit and small debris around the soles, a paper scrap blowing past. Warm sunlight raking across the asphalt texture, long shadows. 16:9, photorealistic, fine film grain in texture only, full-bleed edge-to-edge, no film border, no vignette frame, no light leaks, no text, no watermark.
Keyframe 3: the sprint
What it does: pins the final cut. Real runner biomechanics written into the prompt keep the sprint from going rubbery.
The sprint into the smoke. Grounded stride, blade trailing.
KEYFRAME, SPRINT
Cinematic film still, sprint keyframe. Setting: exactly the bright Manhattan street from the first reference image, this stretch filling with drifting golden smoke, the towering backlit dust wall close, parked cars fading into haze. SIDE VIEW profile, 35mm handheld at chest height: the robot soldier from the second reference image in the SECOND STRIDE of an accelerating sprint, moving left-to-right into the smoke, with NATURAL HUMAN RUNNER BIOMECHANICS: torso leaned forward about 25 degrees, rear foot planted toe-off with small sparks, front leg swinging low with knee driving forward, both feet close to the ground, arms in natural opposed running swing, the right hand gripping the katana low with the teal blade angled back and down along his stride, trailing a faint glow line. Head level, visor locked forward on the smoke. Grounded, powerful, believable acceleration — NOT a leap, NOT airborne, no extreme lunge. Debris and papers whip past with motion blur, dust curling in his wake, golden rim light on the armor. 16:9, photorealistic, fine film grain in texture only, full-bleed edge-to-edge, no film border, no vignette frame, no light leaks, no text, no watermark.
Stage 2: How to run it, step by step
No video tutorial needed. Do these in order.
- Make your character sheet. Use the template in Stage 1 with your own photos. Save the result. This file comes back in step 5.
- Make your robot sheet and your location plate. Two prompts, two stills. Swap the emblem and the city to make them yours.
- Build the three keyframes. Crowd hook, feet, sprint. Each one uses your plate plus the right sheet as references. If a keyframe is not right, re-roll the still. Never fix stills with video credits.
- Get your video prompt. Copy my final prompt in Stage 3 as is, or hand your AI the brief below with your own references attached and let it write yours.
- Fire it in Higgsfield. Seedance 2.0, 15 seconds, 16:9, genre action. Attach in this order: crowd keyframe as the START frame, then feet, then sprint, then your character sheet. Start at 720p. The ladder below explains why.
THE BRIEF, hand this to your AI with your references
TASK: Write one production-ready video generation prompt for Seedance 2.0 from this brief and the attached reference images. Output a single prompt, ready to paste, nothing else. THE VIDEO (15 seconds, 16:9, one generation): A man walks calmly down a sunlit New York side street while a huge dust cloud from an explosion towers at the far end of the block and a panicked crowd sprints past him the other way. We find him through the fleeing crowd (REF-1 is the opening frame). The camera drops to ground level on his sneakers walking through debris (REF-2). He stops. He kneels down into the frame and places his palm flat on the pavement, fingers spread. The moment his hand touches the asphalt, a transformation ignites from the point of contact: liquid gunmetal robot armor with teal accents spreads from his fingers up his arm and over his whole body while the camera slowly pulls back, until he is a full robot soldier, kneeling, and a helmet with a glowing teal visor seals over his face last. Still kneeling, he reaches over his shoulder and draws a katana from a sheath on his back. Then he looks up and sprints hard toward the dust cloud, into the danger everyone else ran from, and disappears into the smoke (REF-3 shows the sprint). HARD RULES: - 15 seconds total, 16:9, photorealistic, real-time speed, no slow motion - Environmental sound effects only. No music, no voiceover, no subtitles. - No on-screen text, no logos except what exists on his clothing - The man never smiles. Calm, certain, unhurried. - Buildings, street and parked cars stay rigid; only people, dust and debris move.
I ran this exact brief through two models. Claude Fable 5 wrote the prompt you see in Stage 3, and GPT 5.6 Sol wrote the one below. Same brief, same four reference images, two completely different directing styles: Fable cut the film into three shots, 5.6 directed it as one continuous unbroken take. Copy whichever style fits you, or run the brief through your own AI and get a third.
THE GPT 5.6 SOL VERSION, same brief, one continuous shot
SEEDANCE 2.0 — 15 seconds — 16:9 — one continuous unbroken shot — real-time speed — photorealistic cinematic action. REFERENCE ROLES: @image1 establishes the opening composition, New York location, fleeing crowd, warm backlight and towering dust cloud. @image2 defines the hero's red-and-black Nike sneakers and the asphalt-level perspective. @image3 defines the final robot silhouette, gunmetal-black segmented armor, teal mechanical accents, glowing teal visor and sprinting physicality. @image4 is the absolute identity and wardrobe reference. Preserve this man's exact facial structure, age, hairline, proportions, black Nike Air T-shirt, blue jeans and sneakers until his face is covered. No identity drift. Begin on the composition of @image1: a compressed telephoto view down a sunlit New York side street. A colossal amber-grey dust cloud boils upward at the far end of the block. Terrified civilians surge toward the camera, fleeing the cloud, crossing close to the lens in chaotic layers. Between their moving bodies, reveal the hero walking calmly toward camera. He is the only composed person in the scene—expression controlled, jaw set, never smiling, gaze unwavering. His pace is measured while the crowd rushes around him. The camera retreats ahead of him, then slips past a fleeing pedestrian in a motivated foreground wipe and dives smoothly to ankle height. Now track backward inches above the asphalt, framing the red-and-black sneakers from @image2 stepping through grit, loose paper and tiny fragments of debris. Natural footfalls, convincing weight and contact. After two deliberate steps, both shoes stop dead. Without cutting, the camera remains low as the hero kneels into frame. One knee strikes the pavement. He plants one bare palm flat against the asphalt, fingers spread. At the exact instant skin touches ground, a sharp teal pulse flashes beneath his hand. Reflective liquid gunmetal erupts from the contact point—not smoke and not fabric—flowing over each finger as articulated mechanical plating, climbing his wrist and arm in branching streams of black metal and teal light. The camera performs a fast, elegant 180-degree arc around him while gradually pulling back, revealing the transformation traveling continuously across his shoulders, torso, back and legs. Clothing is replaced by interlocking gunmetal-black armor plates with exposed mechanical joints and restrained teal accents. Preserve realistic mass, engineering and physical continuity: no body duplication, no melting anatomy, no floating pieces. A katana sheath assembles securely across his back as part of the armor. His human face remains fully visible and identical to @image4 until the final moment. Then cheek plates close, the helmet seals around his head, and a narrow teal visor ignites last. Still kneeling, the completed robot reaches over one shoulder and draws the katana in one clean metallic motion. The blade catches the warm sunlight with a subtle teal edge. He raises his helmet toward the distant cloud for one decisive beat. He explodes into a full-power sprint away from camera, directly toward the disaster everyone else is escaping. The camera drops even lower and accelerates beside him into a dynamic three-quarter side-tracking angle inspired by @image3. Mechanical feet strike the road with tremendous weight, throwing sparks and grit; armor pistons articulate naturally; the katana stays controlled in one hand. The fleeing crowd parts behind him as he charges toward the towering dust wall. End with his armored silhouette swallowed completely by the turbulent smoke while the teal visor remains visible for one final fraction of a second, then disappears. VISUAL LANGUAGE: premium large-format action cinematography, grounded superhero realism, warm late-afternoon sunlight against cool teal armor illumination, deep dimensional atmosphere, realistic exposure shifts, subtle anamorphic character, controlled natural motion blur, tactile asphalt and metal, crisp subject detail, dramatic scale. The camera movement must feel physically operated and spatially continuous—no cuts, no teleporting, no time jumps, no slow motion. ENVIRONMENTAL CONTINUITY: brick buildings, fire escapes, streetlights, pavement and parked cars remain perfectly rigid and stationary. Only people, airborne dust, loose paper, grit and small debris move. The dust cloud remains at the far end of the same block throughout. AUDIO: environmental sound effects only—panicked footsteps, distant shouts without intelligible dialogue, turbulent dust, fluttering paper, shoe impacts, knee impact, rising liquid-metal transformation, locking armor plates, visor activation, katana unsheathing, heavy mechanical sprint and debris impacts. No music. No voiceover. NEGATIVE CONSTRAINTS: no subtitles, captions, interface graphics, watermarks or film borders; no added logos or robot emblems; no smiling; no facial mutation; no identity drift; no extra limbs or fingers; no fused crowd bodies; no rubbery armor; no fantasy magic particles; no new explosion during the shot; no moving cars or buildings; no camera cuts; no freeze frames.
Stage 3: The final prompt
This is the exact prompt behind the video at the top of the page, written by Claude Fable 5 from the brief above. One generation, four cuts inside it, all directed in text.
THE FINAL VIDEO PROMPT, Fable 5
Style: 8K IMAX. Photorealistic — no 3D render, no game engine. Lighting: Bright crisp late-morning sunlight raking the street, vivid blue sky, the dust shockwave brilliantly BACKLIT glowing cream-gold. Teal glows (chest logo charge, armor seams, pavement energy veins, visor, blade) are diegetic emitted light — the only cool light in a warm golden world. No other artificial lighting. Color: 60:30:10 — dominant sunlit warm brick and cream dust / secondary neutral asphalt and gunmetal / accent teal (glow only). Camera: Physical cine lens. 180° shutter motion blur. 16:9 widescreen. Never locked off — each cut rides one continuous motivated move. Skin: Pore-level realism — vellus hair, capillary flush, pore-shadow matching sunlight. Acting: Hollywood — micro-pauses, precise eye-lines, living eyes, chest rise from breathing. Every person shows readable emotion and moves from frame one — no statues. CAL NEVER SMILES — his register is calm, jaw-set certainty. Physics: Gravity and inertia respected — real weight, correct contact shadows, no floating props. Streets, buildings and parked cars stay stationary and rigid; only people, dust, papers and debris move. Continuity: Characters, props, environment identical across every cut. No identity drift. The four reference frames are the visual law — each phase matches its frame. Technical: 24fps smooth motion at REAL-TIME CONSTANT SPEED throughout — absolutely no slow motion, no speed ramps, no time remapping in any cut. No jitter. No timestamp overlay, no camcorder overlay, no VHS artifacts, no film borders, no light leaks, no watermark, no on-screen text. Audio: Environmental SFX only — no music, no subtitles, no voiceover. Characters: FACE LAW: the multi-view character sheet on the gray studio background is the only face reference for CAL — his face matches it exactly in every frame, no identity drift. CAL — the calm man from the reference frames: mid-40s, short salt-and-pepper hair, black t-shirt with white Nike Air logo, dark blue jeans, black-and-red sneakers. Alive and reactive, NEVER smiling. CM ROBOT SOLDIER — the robot from the sprint reference frame: gunmetal-black segmented armor, teal accents, metallic CM emblem on the teal chest plate, glowing teal visor, articulated mechanical hands and feet, slim scabbard flat across his back, katana with teal-burning edge. He IS Cal — same build. Scene: A sunlit Manhattan side street, brilliant blue sky, a huge backlit cream-gold dust shockwave towering at the far end. A panicked crowd flees the dust; Cal walks calmly among them — until he stops, kneels, touches the pavement, and the machine takes him. Then he runs TOWARD the storm. Architecture and parked cars stay rigid throughout. CUT 1 (0.0s–3.0s) — THE HOOK — match the reference frame of the man seen through the fleeing crowd. 135mm long lens at shoulder height, camera embedded in the crowd, gliding backward with the flow, jostled handheld energy: Panicked pedestrians rush past close to the lens on both sides, shoulders and whipping hair smearing the frame edges with heavy motion blur, faces terrified, shouting. Through the moving gaps, in crisp long-lens focus: CAL walking steadily toward camera, the only calm figure, eyes alive and level, golden dust wall towering behind him. SFX: rumble, screams, alarms, pounding footsteps close to the mic. CUT 2 (3.0s–12.0s) — THE ONER — one single unbroken shot, 35mm, lens inches above the asphalt. Opens matching the ground-level reference frame of the walking Nike sneakers, then evolves CONTINUOUSLY within the same take — no dissolves, no crossfades, no morph-cuts, no re-framing jumps anywhere inside this shot: Phase 1 (3.0s–5.0s): at normal real-time walking speed, the black-and-red Nike sneakers walk toward the low camera through grit and skidding papers, blurred fleeing legs deep in background — then they STOP. A beat of stillness. Phase 2 (5.0s–7.0s): he drops to one knee INTO frame — torso, then face entering the shot, eyes down, hard focus, no smile. His right hand comes down: PALM FLAT on the pavement, FINGERS SPREAD WIDE. One breath. Phase 3 (7.0s–12.0s): on contact, ignition — his fingers and hand turn to segmented gunmetal metal, the transformation crawling visibly up the arm exactly as in the kneeling reference frame, a teal energy pulse rippling outward through the asphalt cracks in glowing veins, the printed Nike swoosh on his chest catching a teal charge and lighting up. The camera begins a SLOW CONTINUOUS PULLBACK, widening as liquid gunmetal floods over his shoulders, torso, legs — the helmet sealing over his face LAST, visor snapping alight. Fully machine, still kneeling, head bowed, his right hand reaches OVER his right shoulder, grips the katana hilt protruding behind his back from the scabbard lying flat across his back, and draws the blade UP and OUTWARD in one wide arc away from his body — the blade clearing the scabbard behind him, never near his neck — then settles low at his side, teal edge ringing. SFX: footsteps stop, one breath, palm slap on asphalt, deep sub-bass pulse, rising metallic cascade of plates locking, servo whines, the blade's ring. CUT 3 (12.0s–15.0s) — THE SPRINT — match the side-view sprint reference frame. 35mm handheld profile at chest height, slight dutch, tracking with him: His helmeted head SNAPS UP, visor blazing — then he accelerates into a sprint toward the smoke, moving left-to-right with NATURAL HUMAN RUNNER BIOMECHANICS at real-time speed: torso leaning about 25 degrees, feet low to the ground, each stride driving off the rear toe with sparks, arms in opposed running swing, weight and inertia believable. The katana stays in his right hand, held low and TRAILING BEHIND his body, blade always clear of his legs, drawing a teal line through the haze. Debris and papers whip past with motion blur, the handheld camera straining to keep up. In the final frames he outruns the lens and vanishes into the golden dust. SFX: one heavy breath inside the helmet, pounding metal footfalls accelerating, wind roar, debris.
The anti drift trick. Attach your character sheet as the last reference on the video generation. The prompt names it as the face law. Without it, faces and logos slowly melt between cuts.
The resolution ladder
Same prompt, same references, three fires. 720p first as the cheap proof at 67 credits, judge the motion. Then 1080p at 135 to check that faces and logos hold. Then 4K on high bitrate at 330 for the master. Debugging at 4K prices is how credits die. My final path, stills included, came in around 565 credits.
Recreate it with your own face
- Shoot four to six photos of yourself in the clothes you want on camera and run the character sheet template.
- Decide what you transform into. Robot, knight, astronaut. Two-view sheet, front and back.
- Pick your street and generate the plate. Your city works better than mine, it is yours.
- Build the three keyframes from the plate plus your sheets, and approve them as stills.
- Fire the final prompt at 720p, climb the ladder, post the 4K. Tag me.
Everything on this page is free. The pack has the exact reference images, the robot sheet, the location plate, and every prompt as a numbered text file, including the character sheet template.
Download the Workflow Pack zip, 54 MB: references, build assets, 10 prompt files including both AI versions, README
If you build one, tag @calsaistudio so I see it. More free workflows: Corona, Jack Daniel’s, One Final Kiss. And if you want this built for your brand instead of learning it, that is literally my job.
