- Stack
- Higgsfield (Seedream 5 Pro, Seedance 2.5), ElevenLabs, any editor
- Workflow
- 8 steps · 7 prompts
- Models
- Seedream 5 Pro Seedance 2.5 ElevenLabs
- Keyword
- Comment GTA under the reel ↗
- Price
- Free

What you need
One face sheet of your character · Higgsfield (Seedream 5 Pro for the sheets, places and start frames, Seedance 2.5 for every video) · ElevenLabs for the hook voice · any editor.
08 stepsThe workflow
- 01
One character sheet per outfit
From your face sheet, make one sheet per look: 4 face angles, 6 expressions, hands, fabric, 3 full-body views, hair down. Put a sheet in the references of every image and every video: that is what keeps the same face across five outfits.
- 02
Empty places first
Generate each place with nobody in it, from the exact camera angle you want: gas station at night, art-deco avenue at sunset, turquoise islands, foggy swamp, rooftop. Then the start frame = outfit sheet + empty place + camera angle. It looks far better than asking for the character and the place at once.
- 03
Real night light, not neon everywhere
Magenta and cyan neon on wet asphalt is the most overused AI look. Ask for orange sodium streetlights, flat white fluorescent tubes and one single neon sign, the rest dark. Same for day: one low sun, haze, true colors.
- 04
She never poses
Catch her in the middle of something (eating chips at the pump, wringing her wet hair), off-center, not looking at the lens, with strangers walking in front of the camera. Posing to the camera is what makes it look generated.
- 05
Film every crime as found footage
The joke of each scene is filmed by someone else: a ceiling security camera, a bystander's phone, a police body cam, an action cam on the boat, a news helicopter, a mugshot. Ugly, shaky, real: the AI defects disappear in it.
- 06
Seedance for every clip
Start image + outfit sheet, 720p. The outfit clip is 3 shots in 6 s with hard cuts (wide, then two close-ups of the clothes). The crime clip is one continuous take of 5 s with its sound (shouts, sirens, engines). Cut each shot to 1 or 2 seconds in the edit.
- 07
Anything that drives away needs an end frame
For a chase, leave open water or road between the camera and the vehicle in the start frame, and add an end frame with the vehicle far away. Without it, Seedance can make it drive backwards.
- 08
The hook
Write the voice first in ElevenLabs, then give it to Seedance as the audio reference with a selfie start frame. Make it look re-uploaded (720p, low bitrate) and add a comment-reply bubble with the question you are answering.
!Avoid this
Never name the game, a character or a brand in a prompt: describe the place instead, or the image gets blocked. No neon everywhere, no posing, no long shots. No white flashes between scenes: straight cuts only.
READY TO COPY · SEEDREAM 5 PRO + SEEDANCE 2.5 + ELEVENLABS
The prompts
Replace the words in [BRACKETS] with your own character, outfits and places.
Outfit sheet
Character sheet of [YOUR CHARACTER], a photorealistic young woman, on a light grey seamless studio background, soft studio light, editorial fashion photography. Use the reference image as her exact face. HAIR: [e.g. long straight black hair worn LOOSE, two blonde face-framing streaks]. OUTFIT: [e.g. an oversized vintage black bowling shirt with a wide red panel across the chest, open over a white ribbed tank top; baggy black cargo pants with a silver wallet chain; chunky black sneakers]. LAYOUT: top row FACE FRONT, FACE 3/4 LEFT, FACE 3/4 RIGHT, FACE PROFILE; second row EYES CLOSED, DEADPAN, SMIRK, LAUGHING, SURPRISED, [ONE MORE EXPRESSION]; third row HANDS DETAIL, FABRIC DETAIL; bottom row FULL BODY FRONT, FULL BODY 3/4, FULL BODY BACK. The same outfit in every panel, a clean label under each panel. AVOID: no logos, no brand names, no text on clothing, no tied hair.
Empty place
CAMERA: [e.g. long lens from across the street, slightly low angle, compressed perspective]. PLACE: [e.g. a run-down gulf-coast town at night: a 24-hour gas station with a flat canopy, old fuel pumps, a small convenience store, power lines, palm trees, a cracked damp parking lot, a rusty pickup at a pump]. The spot where she will stand stays empty. LIGHT: [e.g. orange sodium streetlights, flat white fluorescent tubes with one flickering, a single pink neon sign, the rest dark]. WORLD: hyper-detailed and lived-in: faded paint, rust, puddles, a little litter, a few people in the background going about their lives. STYLE: photorealistic film still, landscape 4:3. AVOID: no text, no logos, no brand names, no person in the foreground, no video-game look, no waxy CGI.
Start frame
REFERENCES: the character sheet is [HER] exact face, hair, jewelry and outfit; the place image is the exact place, light and camera angle. SHOT: [e.g. candid photo taken from across the street with a long lens, handheld: she leans against a gas pump in the right third of the frame eating chips straight from the bag and laughing at someone off-frame; a man walks across the foreground, motion-blurred]. She is not looking at the camera. LIGHT: same as the place. STYLE: candid documentary photo, caught in the moment, true colors, a little sensor noise. AVOID: no posing, no looking into the lens, no centered composition, no text, no logos.
Outfit clip
START FRAME: the video starts exactly on the start image. Use the character sheet only to keep her face, hair and outfit identical. THREE SHOTS with hard cuts. SHOT 1 (0 to 2.6 s): the start image, handheld: [e.g. the man walks out of frame, she laughs with her mouth full, bending forward]. SHOT 2 (2.6 to 4.2 s): [e.g. ground-level close-up: her sneaker steps into a puddle]. SHOT 3 (4.2 to 6 s): [e.g. close-up of her hand brushing chip dust off the red panel of her shirt]. LIGHT: [same as the start image]. Real handheld footage, natural motion blur, real-time speed, not slow motion. AUDIO: [e.g. pump click, fluorescent hum, distant traffic, her laugh]. No music. AVOID: no posing, no looking into the lens, no text, no change of outfit or face.
Crime as found footage
START FRAME: the video starts exactly on the start image. Use the character sheet only to keep her face, hair and outfit identical. CAMERA: [e.g. shaky phone video filmed by a bystander in the parking lot: the person filming flinches as she runs past, the phone swings and loses her for a moment, autofocus hunting, auto exposure pumping]. No cuts. ACTION: [e.g. she sprints past laughing, clutching snacks and a slushie; the cashier runs out after her shouting, then stops, out of breath, hands on his knees]. Real-time speed, heavy natural movements, nothing graceful or posed. AUDIO: real phone-mic sound: [e.g. the cashier shouting "Hey! Get back here!", her laugh, crinkling bags, the person filming saying "oh my god"]. No music. AVOID: no slow motion, no smooth camera, no text, no logos.
Hook voice
[casual, amused, reading her phone] dress like one? [soft scoff] easy. [short pause] [playful, a little lower] act like one... [smirking] that's illegal.
Hook selfie
START FRAME: the video starts exactly on the start image. CAMERA: handheld front-camera selfie at arm's length: small natural hand shake, tiny reframes, auto-exposure breathing, no cuts. ACTION: she starts talking right away, glancing between the screen and the lens; on "easy" a small shrug; on "act like one..." one raised eyebrow; on "that's illegal" a slow smirk; then she sips her drink and winks. VOICE: the reference audio is her exact voice and line; her lips match it precisely. CAMERA QUALITY: slightly degraded iPhone front-camera footage, neutral colors, window daylight. AUDIO: only her voice, close phone-mic sound. AVOID: no phone visible, no text, no subtitles.
The ending
Build up to the last scene: rooftop, police helicopter searchlight, arms wide open. Then cut to the mugshot: one flash, a smirk, a wink. That is the punchline, and it loops back to the selfie.
@LEARNWITHKLOE · TUTORIAL · Higgsfield (Seedream 5 Pro, Seedance 2.5), ElevenLabs, any editor