A 16-block prompt template, measured
Cully Hill Boys follows a fixed 16-block prompt template in which the reference is split into what to use and what to ignore, and where restraints are named and stated in the positive. The measurements come from a sample of the stored prompts, not the full history.
This is a reverse-engineering of the template behind the prompts of Cully Hill Boys, measured across 5,279 complete prompts (2,995 of them use the template).
A note on coverage: the stored prompts are a sample, not the full history of the project. The sample is 5 prompts per scene plus anchor prompts from the corpus. The percentages below are real presence in that sample.
Everything below is measured, not assumed.
The skeleton
Sixteen blocks, each with an UPPERCASE header on its own line, always in this order. The percentage is how many of the prompts that use the template include that block.
| # | Block | Presence | Weight | What it handles |
|---|---|---|---|---|
| 1 | SCENE CONTEXT |
100% | 5% | INT/EXT, time of day, what happens, how many shots, duration |
| 2 | ACTIVE REFERENCES |
93% | 16% | Each reference with what to take and what to ignore from it |
| 3 | LOCATION MAP |
58% | 6% | Where each body is, in frame coordinates |
| 4 | FIRST FRAME AND SPATIAL BLOCKING |
67% | 5% | The frozen first frame, position by position |
| 5 | FORMAT MODE |
69% | 2% | How many shots, how many seconds each, where the cuts fall |
| 6 | OPTICS |
74% | 3% | Focal length and field of view |
| 7 | CAMERA |
78% | 4% | What the camera does in each beat |
| 8 | ACTION TIMING |
69% | 13% | The timed action, span by span |
| 9 | CHARACTER ACTING |
15% | 6% | Performance per character, one per line |
| 10 | PHYSICS |
75% | 4% | Weight, inertia, cloth, breathing |
| 11 | LIGHTING |
85% | 4% | Source, direction, contrast, grade |
| 12 | AUDIO |
90% | 4% | Diegetic sound, literal dialogue, voice per character, negative close |
| 13 | STYLE |
77% | 3% | Author references and general texture |
| 14 | QUALITY |
76% | 2% | Camera, optics, grain, shutter |
| 15 | POSITIVE LOCKS |
19% | 13% | What cannot move, stated in the positive |
| 16 | POSITIVE CONSTRAINTS |
56% | 12% | The desired state described, instead of a list of prohibitions |
The real core is two blocks: SCENE CONTEXT (100%) and ACTIVE REFERENCES (93%). The rest switch on depending on what the shot needs.
A third of the prompt is the two constraint blocks (POSITIVE LOCKS + POSITIVE CONSTRAINTS, 25% combined) plus ACTION TIMING (13%). The work is there, not in the scene description: SCENE CONTEXT is only 5%.
The writing rules
Conventions that repeat inside the blocks, with measured presence:
| Convention | Presence | Times/prompt | Real example |
|---|---|---|---|
<<<uuid>>> inline |
96% | 16.9 | Every mention of a character or object carries its ID |
| Optics in degrees and mm | 81% | 5.9 | ~47° (≈50mm) settling to ~50° |
100% match |
64% | 3.4 | USE FOR: face and body identity, 100% match |
HARD CUT declared |
56% | 3.7 | The cut is named, not deduced |
| Time range | 48% | 4.9 | 4.0–7.0s — Cal lifts the phone… |
| Negative audio close | 46% | 1.4 | No music, no other voices, no subtitles. |
Render EXACTLY ONE |
24% | 2.2 | Quantity control, against duplicates |
| Frame coordinates | 19% | 7.5 | x≈42%, y≈68% |
Line verbatim |
14% | 1.6 | Dialogue goes literal and is marked as such |
IGNORE: in references |
11% | 3.2 | IGNORE the reference's pose, background, lighting and camera angle |
USE FOR: in references |
9% | 3.4 | Which aspect of the reference is used |
Voice for |
5% | 1.5 | Voice casting described per character |
Two of these deserve their own note because they are the system's signature:
The reference is declared in parts, not whole. USE FOR: face, body identity and wardrobe + IGNORE: the reference's pose, background, lighting and camera angle. Never "use this image". Always what to take and what to discard.
The frame is addressed in percentages. x≈42%, y≈68%, with the convention declared in the prompt itself (screen-X 0% left→100% right). It appears in 19% of prompts but 7.5 times in each one: when it is used, it is used for everything.
The template, ready to use
The placeholders below were written in Spanish in the source analysis and are translated here.
SCENE CONTEXT
INT./EXT. <place> — <moment>. <What happens, in one or two sentences>.
<N> shots, ~<X>s. <Continuity note with previous shots, if applicable>.
ACTIVE REFERENCES
<<<id>>> — <Name>: <physical and wardrobe description>. USE FOR: <what to take>.
100% matches the reference. Render EXACTLY ONE <Name>. IGNORE: the reference's
pose, background, lighting and camera angle.
<<<id>>> — <Location>: <geometry and materials>. USE FOR: geometry, materials,
layout and practical light only — not camera angle, framing, added people.
LOCATION MAP
<Description of the geography>. (screen-X 0% left→100% right; screen-Y 0% top→100% bottom)
<Character> at x≈<N>%, y≈<N>%, <depth plane>. <Next>…
FIRST FRAME AND SPATIAL BLOCKING
OPENS on <description of the frozen first frame>, position by position. Then HARD CUT in.
FORMAT MODE
<CONTROLLED MULTI-SHOT SEQUENCE / SINGLE TAKE>. <X>s total, <N> shots:
SHOT 1 = <X>s, SHOT 2 = <X>s. <N> HARD CUTs. No music, no subtitles.
OPTICS
~<N>° (≈<N>mm). Lens locked.
CAMERA
Beat 0 (0.0–<X>s): <what it does>. Beat 1: <what it does>. <Eyeline/angle rule>.
ACTION TIMING
0.0–<X>s — <action>.
<X>–<Y>s — <action, with the dialogue if there is any>.
CHARACTER ACTING
<<<id>>>: <performance, tics, gaze, line delivery>.
<<<id>>>: <same>.
PHYSICS
<Weight, inertia, cloth, breathing>. Natural 180-degree shutter cadence.
LIGHTING
Scene-type: <type>. <Source and direction>. <Contrast and falloff>. <Grade and saturation>.
Keep it <consistency> in every frame — do not drift.
AUDIO
Diegetic — <ambience>. Line verbatim:
<CHARACTER> (<intent>): "<exact line>"
Voice for <<<id>>>: "<age, timbre, register, accent>"
No music, no other voices, no subtitles.
STYLE
<Genre and author references>. <Texture>.
QUALITY
<Camera and lenses>, <grain>, <shutter>, <tint>, <era>. No modern elements.
POSITIVE LOCKS
<What cannot move, stated as fact: duration, fps, color temperature,
which character is where and for how long>.
POSITIVE CONSTRAINTS
<The desired state described as if it already existed: "The blank stare is the engine
of the scene: total, dumb, unreadable stillness". Never a list of NOTs.>
What makes this system different
- It names its constraints.
HEIGHT LOCK,WALL ANCHOR,ARM LOCK. A rule with a proper name can be reinforced later without rewriting the paragraph. - It splits every reference into USE FOR / IGNORE. The model copies the pose, background and light of the reference unless explicitly told not to.
- It declares when the visual reference does not rule.
overrides any impression from stillsappeared 8 times in the corpus and was never removed. - It phrases things in the positive.
POSITIVE CONSTRAINTSis in 56% of prompts and was the block that grew most between June and July (30% → 45%). Describing the desired state instead of listing prohibitions appears to be where they ended up.
Limitations
The project's time window is short (June to August 2026) and August has only 52 prompts, so the month-by-month evolution is readable only between June and July. The percentages come from the stored prompts, which are a sample of 5 per scene plus the corpus anchors, not the project's total. The template also coexists with prompts in Chinese that have a different structure (风格与氛围： / 音效：), which fall outside this analysis.