Pressure is a privilege.
An archer, a drawn bow and a tear. You feel the tension before the explanation.
How we made a realistic AI reel, step by step. Tools, prompts and process.
Jump to the prompts
An archer, a drawn bow and a tear. You feel the tension before the explanation.
She's not real. But you still felt it. That is the reveal.
See the same realism in candid AI UGC, then try the workflow yourself.
Find the tension, then write the reveal. Map the hook → twist → proof before generating anything.
View brief / prompt ↗Build three archer shots and six candid UGC stills. Lock composition before motion.
View brief / prompt ↗Animate three close-ups with slow zooms, precise tears and minimal sound.
View brief / prompt ↗Animate each still with small, believable movements. Keep about 1.2 seconds per clip.
View brief / prompt ↗Direct one consistent voice, line by line. Let pauses carry the twist.
View brief / prompt ↗Cut to real word timestamps. Add subtitles, stamps, music ducking and sound effects.
View brief / prompt ↗Scripts, direction, prompts and iteration.
Archer references and six UGC images.
Image-to-video for the hook and proof clips.
Adam · Engaging, Friendly and Bright. American English. No Effect. Stability 0.4–0.5.
Via Claude Code. Motion graphics and a reusable edit template.
Build the page and support the production workflow.
Choose a stage below. Each block names its model. REPLACE marks a final prompt not yet supplied.
Reference images · GPT Image
Three shots, in order: wide → eye close-up → eyes, nose and mouth (5:4 crop). Describe the composition explicitly. The string passes BEHIND the fingers. Add the crowd in layers, then subtle light in separate edits.
Final archer image prompt not supplied. Replace with the approved prompt for this shot.
Model: GPT Image
Final archer image prompt not supplied. Replace with the approved prompt for this shot.
Model: GPT Image
Final archer image prompt not supplied. Replace with the approved prompt for this shot.
Model: GPT Image
Hook video · Seedance 2.5
5 seconds total. Use 01_plano_general, 02_ojo and 03_ojos_nariz_boca in order, for about 1.5 s, 2 s and 1.5 s. Slow progressive zoom in on each shot. The final prompt should describe the tear forming and sliding slowly, mouth and string tension, tense minimal sound, and no morphing, text or fast cuts.
Final Seedance hook prompt not supplied. Replace with the approved image-to-video prompt.
Model: Seedance 2.5
UGC reference images · GPT Image
For all six: attach the reference only for composition and mood. Add “new person, do not copy the face or identity of the reference”.
Vertical 9:16 photo taken from directly above, a young man lying on his back on wide concrete steps, head at the bottom of the frame, looking up into the camera. Black oversized jacket with metal toggle closures, black trousers, black shoes. White wired earphones plugged into a phone resting on his chest, the cable running over his jacket. Dusk light, on-camera flash on his face, slightly overexposed skin with visible pores, 35mm film grain, muted warm tones. Casual snapshot, not posed. New person, do not copy the face of the reference.
Model: GPT Image · 9:16
Vertical 9:16 flash photo inside a subway car, a young man holding a steel pole, looking away to the side with a calm serious expression. Black cap worn backwards with a small yellow logo, cream sweatshirt with a faded college-style print, thin chain necklace, one white wired earphone, white backpack strap. Fluorescent ceiling lights with a greenish-yellow cast, direct on-camera flash, hard shadow on the ceiling behind him, real subway scratches and reflections. Candid street-photo look. New person, do not copy the face of the reference.
Model: GPT Image · 9:16
Vertical 9:16 selfie taken with a phone held low and close, a young woman with slicked-back hair in a bun, her chin and mouth hidden in the collar of a large faux-fur tiger-striped coat. Looking straight into the lens. Paris street, Haussmann buildings behind her, overcast white sky, a red restaurant sign at the top corner. Front-camera phone look, slight wide-angle distortion on the fur, soft natural light, real skin texture, flyaway hairs. New person, do not copy the face of the reference.
Model: GPT Image · 9:16
Vertical 9:16 film-flash photo, a young woman with two loose braids standing in a small kitchen, holding a white instant camera in front of her face, eyes closed, mouth slightly open as if laughing. Light-blue shiny track jacket zipped up and denim shorts, rings on her fingers. Old white fridge with magnets and photos behind her, a checked kitchen towel hanging, a window with green leaves. On-camera flash, lived-in kitchen, 35mm film grain. New person, do not copy the face of the reference.
Model: GPT Image · 9:16
Final still-image prompt not supplied. Replace with the approved prompt.
Model: GPT Image · 9:16
Final still-image prompt not supplied. Replace with the approved prompt.
Model: GPT Image · 9:16
6 animations · Seedance 2.5
Generate 4 seconds per clip at 720p, 9:16. Use about 1.2 seconds of each in the edit. The Copy buttons automatically include the common block in each animation prompt.
Phone-recorded look, slight handheld shake, natural imperfect light. Keep the face, outfit and background exactly as in the image. No zoom, no camera cuts, no morphing, no text, no extra fingers, hands keep a stable natural shape.
Model: Seedance 2.5
The woman slowly chews the fry in her mouth, her eyes move from the food to the camera. Her hand holds another fry over the box and dips it, small natural movement. The bed pillow shifts slightly as she breathes. [common block]
Model: Seedance 2.5 · 4 s · 720p · 9:16
The boy slowly turns his head from looking left toward the camera and gives a small blink. The earphone cable sways with the movement of the train, the whole car rocks gently. Subtle handheld shake. [common block]
Model: Seedance 2.5 · 4 s · 720p · 9:16
The woman lowers the fur collar slightly with her chin, shows her lips for a moment, blinks slowly and looks into the lens. A few loose hairs move in the wind. Phone selfie micro-shake from her arm. [common block]
Model: Seedance 2.5 · 4 s · 720p · 9:16
The man lying on the steps scrolls the phone with his thumb, his eyes move along the screen, then glance up toward the camera. The earphone cable moves slightly with his breathing. [common block]
Model: Seedance 2.5 · 4 s · 720p · 9:16
The woman presses the shutter of the instant camera, a quick white flash fires, then she lowers the camera and laughs, her braid swaying. [common block]
Model: Seedance 2.5 · 4 s · 720p · 9:16
The boy on the left presses a button on the camcorder and holds it steady at the lens. The boy with the clear glasses blinks once and tilts his head very slightly, serious expression, while a bus passes behind them on the street. [common block]
Model: Seedance 2.5 · 4 s · 720p · 9:16
Artlist · Eleven v4
Voice: Adam, Engaging, Friendly and Bright. American English · No Effect · Stability 0.4 (working range: 0.4–0.5).
A confident, warm American man in his late 20s, speaking like a smart friend telling you something surprising. Natural, conversational, never flat or robotic, with small smiles you can hear. The delivery changes line by line:
Line 1 ("Actually... she feels no pressure."): calm, relaxed, slightly amused.
Line 2 ("She's not real."): a tiny smile, almost a soft chuckle. Then a clear pause.
Line 3 ("But you still felt it."): slower, lower and sincere, stretching "felt it" like a quiet realization.
Line 4 ("Most people can't tell anymore... if it's AI or not."): faster and casual, like sharing a funny secret with disbelief, a short light laugh on "anymore".
Line 5 ("The UGC wave is here!"): upbeat and energetic, big smile in the voice, rising energy on "here".
Line 6 ("Brands are already starting to use it."): fast, confident and conversational.
Line 7 ("And the ones who jump in now... will get a head start."): speed up through "jump in now", then slow down and emphasize "head start".
Line 8 ("Wait too long... and you'll fall behind."): lower and slower than everything else, a knowing half-smile, friendly advice and not a threat. Pause after "long", let the pitch fall on "behind".
Line 9 ("Comment MODEL, and I'll show you how."): warm, friendly and inviting, clear smile, soft rising tone on "how".
Leave a natural pause between lines 3 and 4, and between lines 7 and 8.Model: Eleven v4 via Artlist
Actually... she feels no pressure. She's not real. ... But you still felt it. ... Most people can't tell anymore... if it's AI or not. The UGC wave is here! Brands are already starting to use it. And the ones who jump in now... will get a head start. ... Wait too long... and you'll fall behind. Comment MODEL, and I'll show you how.
Model: Eleven v4 via Artlist
Edit brief · Claude Code + Remotion
1080x1920, 30 fps, ~31 s; hook 6.4 s -> freeze and shrink into a card on cream -> word-by-word subtitles with one accent color -> montage of the 6 clips (about 1.2 s each, TikTok-style frame, "100% AI" stamp) -> text cards for the opportunity -> CTA "Comment MODEL" + 6teen @sixteen.ai. Timings driven by real word timestamps of the voice-over; music ducked under the voice; SFX on cuts and stamps.
Model: Claude Code + Remotion
The full edit prompt is available on request.
Avoid words like "cinematic, epic, teal, haze, glow": they trigger the AI look. Describe real, boring light instead.
Edit images in small steps (crowd first, then subtle light) instead of one big edit.
Hands are the hardest part. Describe exactly where the fingers are and what passes behind them.
Describe the composition explicitly. A reference image on its own gets ignored.
Never copy a real person's identity from a reference: ask for a new person.
Use short clips (1-1.5 s) with minimal movement for proof footage; it hides AI faults and keeps retention.
One consistent voice for all explanations; avatars can keep their own voices.
A hybrid format (about 40% realistic, 60% motion graphics) cuts cost a lot and keeps the brand consistent.
A qualitative breakdown of this workflow, without invented prices.
| Part of the process | Relative cost & effort | What carries forward |
|---|---|---|
| Script & direction | Low asset cost; takes creative iteration. | Hook → twist → proof structure. |
| Reference images | Repeated edits take time, especially hands. | Approved stills and composition prompts. |
| Realistic video | The expensive part to iterate. Keep movement minimal. | Prompt patterns and selected short clips. |
| Voice-over | Delivery and pauses need careful listening. | One consistent voice and direction block. |
| Motion graphics | Upfront setup; cheaper to reuse across reels. | Remotion template, stamps and subtitles. |
The mix: about 40% realistic footage, 60% motion graphics. Less generation, more brand consistency.
Follow @sixteen.ai for more workflows and prompts.
Follow @sixteen.aiComment MODEL on the reel.
All people shown are AI-generated.
← Back to the library