Scripting
What the character does and says.
- Script & directorYou
- Co-directorClaude Opus, ChatGPT Astra
Rough idea
Creator’s process- Write ideas down as they come.Rough drafts lived in a personal notes app. No structure needed.
- Brainstorm with an AI-assistant.Paste the whole idea into ChatGPT or Claude with a bit of direction and work out how things should go.
- End with a revised screenplay.The screenplay then goes into direction (chapter 03) to become the scene prompt.
From notes app to finished prompt
A real rough draft, typos and all, next to the Seedance prompt it became.
Next video Skyline slow zoom in, then camera cut to looking up to the skyline from street level, Car drives over camera, we see the underside. Cut to inside the car, see from windshield, city reflects off, cut to him inside, Looking focused He eyes shift slightly, Cut to various things > Crowds. Food. Signs. Scooters. Market stalls. People moving around at a blur like a timelapse. Last cut, Cut back, we see a glass reflection of the city, main character steps into view and his faces becomes dearly seen through the glass. Cut to behind him looking out from his penthouse over the city skyline as dawn begins to break, zooming out, he says “I know you’re out there, brother” Black
18 seconds · 4:3 · Smooth PS2/PSP-style 3D Use the supplied references <<<image_1>>> <<<image_2>>> <<<image_3>>> for the main character, <<<image_4>>> for his car, <<<image_6>>> <<<image_7>>> the city, and <<<audio_1>>> for his voice. Smooth shaded geometry, simplified textures, soft bloom. Preserve character and car consistency throughout. Nighttime until the final dawn reveal. No music or subtitles. Spoken language: English only. Use the supplied audio reference for the main character's voice, tone, and accent. Chinese text appears only on environmental signage. Background crowd chatter is indistinct, with no intelligible words. 0:00–0:01.5 — Above the city High aerial view looking diagonally down across Guangzhou at night. Slowly push toward the illuminated towers, as if beginning a descent into the city. Distant city hum. 0:01.5–0:03 — Car overhead Hard cut to a camera flush with the road, looking up between impressive skyscrapers. The Jaguar drives directly over the lens, its underside briefly filling the frame. Engine growl and tire rush. 0:03–0:04.5 — Windshield Hard cut to the main character through the moving car’s windshield. Tight framing. Neon buildings reflect across the glass while his focused face remains visible. 0:04.5–0:06 — Inside the car Passenger-side close-up. Gentle road vibration. Passing lights sweep across his face. His eyes shift toward the shops, then forward. 0:06–0:11 — City montage Six rapid hard cuts. Every shot feels personally observed: handheld camera with subtle sways and small framing adjustments, as though someone is holding it while exploring throughout the city. Tight, selective compositions squeeze fragments of the city into the frame. People, objects, and storefront edges partially come in and out of view. Shoulder-level view among crowds crossing the street. People pass close to the lens, briefly obscuring the scene. Close view of sizzling food. Steam rises past the lens as a vendor’s hand enters the frame. Camera tilts slightly upward toward glowing Chinese signs crowded between building edges. Scooters pass close by, briefly filling the frame as the camera follows. A passing glimpse into packed Y2K stalls: electronics, jackets, bags, and toys overlapping within the tight composition. Handheld sidewalk view. Pedestrians blur past in accelerated time-lapse, suggesting the night slipping away. Brief overlapping fragments of chatter, sizzling food, footsteps, and engines build the city’s energy. 0:11–0:13.5 — Reflection reveal Hard cut. The busy soundscape drops to a distant hush. Close view of the penthouse glass from outside, initially filled with the city’s reflection. The main character steps into view behind the glass. Focus shifts to his face, clearly visible beneath faint reflections. The first dawn light touches him. 0:13.5–0:18 — Dawn above the city Hard cut behind the main character inside the penthouse. He stands at the floor-to-ceiling windows, overlooking the city as early dawn breaks. Slowly pull back from behind his shoulders, revealing the room and expansive skyline. The main character says exactly in English, without translation or additional dialogue: “I know you’re out there, brother.” Quiet certainty, matching the supplied voice reference. Finish before the final half-second.
The beats stay the same. The prompt adds timings, a camera position for every cut, sound for each shot, language rules, and delivery for the line.
Actions and dialogue
Prompt frameworkOnce the script goes into direction, a few things stay fixed. Mark them in the draft:
- Exact dialogue. Approved lines stay word for word unless a rewrite is requested.
- Who is speaking. Name the speaker on every line.
- Spoken or internal voice. Internal thoughts play with the character’s lips still.
- Reveals. If a face or object stays hidden until a certain moment, say so.
Direct eye contact, curious deadpan: “Why is it called tomfoolery and not just foolery? Who is Tom and what did he do?” One small explaining gesture: “Imagine being so ridiculous, your ‘foolery’ gets credited for generations.” A brief beat, then absolute deadpan: “I can.”
A delivery note before each line, and the pause written in as a beat.
Pacing and reactions
Creator’s processPrompt frameworkA held look after a line is often where the joke or realisation lands. Write it into the script.
- Mark intentional pauses and keep them through revisions.
- Give reactions their own shot: “Hard cut to a locked face close-up. A brief beat, then absolute deadpan: ‘I can.’”
For the one that started it all, 300 words per minute worked best.
The production grammar
This production grammar breaks it down: a teardown of a 26.5-second streetwear promo, reverse-engineered for its shot logic, camera behaviour, pacing and assembly. Rebuild it with an original character, environment and script, and leave the source’s character, wardrobe, IP and dialogue alone.
A shot-size ladder that follows the argument
- MCUHook
- WideContext
- MCUAnecdote
- MSPayoff
- CUCall to action
Keep the ladder; the exact sizes can change. The ladder is what makes 26 seconds feel like a short film.
Camera behaviour is a vocabulary
| Slow push-in | Leaning in to tell you something. | |
| Handheld sway | Casual, off the cuff. | |
| Lateral drift | Time passing; storytelling. | |
| Locked-off close-up | Stop — this is the point. |
The subject stays on the right third with eye contact straight into the lens, which turns a monologue into a conversation with the viewer. The plaza prompt uses the same grammar: a push-in to open, a wide with sway, a lateral drift for the story, a locked close-up for “I can.”
Editing rhythm
- 6 cuts in 26.5 seconds: an average shot of about 4.4 s, the longest 9 s (context), the shortest 1 s (the call to action).
- Every cut lands on a punch word or a caption change.
- Delivery is deadpan and fast: 2–6 word phrases, short holds, and the occasional dead-air beat for a joke. No empty seconds.
Generate long, cut short
Video models have a minimum generation length of around 4 seconds, so a 1- or 2-second shot can’t be generated directly. Generate 4–6 seconds of coverage per beat and set the exact cut point in the edit.
Download the full production grammar — shot table, visual design versus grammar, a Higgsfield tool map, a reconstruction workflow and a QC checklist.