Prompt craft · proposed structures

Ten series concepts for short generated films

Short films whose subject is a transformation. Each has a real-world anchor, an escalation with a natural end, and a texture register chosen to be as far from generic 3D as possible.

Concepts: 10
Each: 5 beats, 15 seconds
Values per concept: 6

These have not been generated

These are designs, not results. Unlike the hand-drawn VFX set, none of them has been run, so there are no timings, no settings and no output to report — and any of them may turn out to be beyond what current models will hold for fifteen seconds. They are written to the same shape as the set that was run, because that shape is what made it reusable.

How to use one

Pick the concept, then pick one value of its variable. Write the five beats with explicit timestamps.
Append the texture discipline and the negatives to every beat, byte-identical across the series. Holding those constant is what makes six films feel like one body of work.
Change nothing but the variable between films. The moment you improve the arc for film three, you have two series.
Read the anchor as a hard requirement, not a suggestion. Each one is the constraint the concept fails without.

The 10 at a glance

Concept	In one line	The variable
Objects remember what they were	A real object briefly reverts to its earlier life and comes back.	The object, and therefore the life it reverts to
The instruction manual comes alive	Diagram lines lift off a flat sheet, assemble the thing above it, fail, and try again differently.	The document the lines come from
Weather that belongs to one object	A single object carries its own microclimate, and it spills.	The weather, and its host object
The sound becomes visible	Something audible in the room emits a form you can see, and the picture follows the audio.	The sound source
One shot, four decades	A locked-off frame in which the room ages around one object that never leaves.	The room and the surviving object
The map redraws itself	A printed map animates its own route across the real surface it is lying on.	The kind of document, and therefore the kind of journey
Scale collapse	Something tiny becomes something vast without a cut, on a matched texture.	The small thing, and the vast thing it matches
Handwriting outlives the writer	Old handwriting lifts off its surface and re-enacts what it describes.	The written surface
The shadow acts on its own	A real shadow stops matching its object, does something, and returns before it is caught.	The light source that casts it
The recipe assembles itself	Ingredients arrive in order, in stop-motion, and the dish builds itself in one frame.	The dish, and therefore the order

Objects remember what they were

A real object briefly reverts to its earlier life and comes back.

Use it for: Material provenance, sustainability, craft heritage, anything where the story is where a thing came from.
The variable: The object, and therefore the life it reverts to

The arc

Beat	Name	What happens
0-3s	The ordinary	The object sits in a real, unremarkable setting, filmed close and handheld. Nothing suggests anything is about to happen.
3-6s	The slip	One edge of it unravels backwards — a rim becomes a growth ring, a cuff becomes fleece — while the rest of the object stays exactly as it is.
6-10s	The full reversion	The reversion travels through the whole object and it becomes the earlier thing, still sitting in the same modern room, wrong-sized and wrong-lit for it.
10-13s	The place arrives	The room briefly loses to the older place: the hillside, the forest, the foundry floor, bleeding in from the edges of frame.
13-15s	The return	It snaps back to the object on the table, unchanged, one beat of stillness, framing slightly off.

Six values for the variable

a paper cup and a tree ring
a knitted sleeve and a sheep on a hillside
a wooden chair and the branch it was cut from
a glass bottle and a beach
a cotton shirt and a field in wind
a steel spoon and a foundry pour

Texture discipline

The reverted state carries a different medium from the live action: archival film grain, a slightly lower frame rate, faded dye. The modern half stays clean digital. Two registers in one shot is the whole effect.

The anchor

The object never moves position between states, and the light on it never changes direction. Everything else can change.

Negatives

no cuts
no dissolves
no morphing through a white flash
no text
no logos
no music

The reversion is a transformation with a natural stopping point, so it does not need an escalation invented for it. And because the object holds still, the model has a fixed anchor to transform around rather than a moving target.

What to watch for

The temptation is to cut to the older place. The whole idea collapses if you do; the older place has to arrive inside the same frame.

The instruction manual comes alive

Diagram lines lift off a flat sheet, assemble the thing above it, fail, and try again differently.

Use it for: Anything with an assembly step, a process, or a set of parts. Onboarding, DIY, hardware, software architecture.
The variable: The document the lines come from

The arc

Beat	Name	What happens
0-3s	The flat page	The document lies on a real surface under real light. A hand smooths it. The paper's own texture and fold shadows are visible.
3-6s	Lift-off	The line work peels up off the page in order, keeping its exact drawn weight, and hangs in the air above the sheet.
6-10s	Assembly	The lines rotate into an exploded view, then converge and build the object in the air, with arrows and hidden-edge dashes behaving as if the diagram were doing the work.
10-13s	The failure	One part joins wrongly. The assembly holds the mistake for a beat, then comes apart deliberately, not violently.
13-15s	The second try	It reassembles correctly, settles onto the page, and flattens back into printed line work exactly where it started.

Six values for the variable

a flat-pack instruction sheet
a handwritten recipe card
a wiring label inside an appliance
a knitting pattern
an origami crease diagram
a fire-escape plan on a wall

Texture discipline

Technical drawing, stated in specifics: uniform hairline weights, dashed hidden edges, exploded-view arrows, section hatching, no fills, no shading, no perspective cheating.

The anchor

The paper stays in shot and in focus throughout, so there is always a real surface establishing that the lines came from somewhere.

Negatives

no colour beyond the ink of the page
no smooth vector gloss
no 3D shading
no text beyond what is printed on the document
no music

The technical-drawing register is as far from slick advertising CGI as crayon is, and it is much less used. The failure beat is what stops it reading as an advert.

What to watch for

Diagram vocabulary is precise and models blur it. If the hairlines start varying in weight, the register is gone and you are back to generic 3D.

Weather that belongs to one object

A single object carries its own microclimate, and it spills.

Use it for: Temperature, freshness, comfort, energy. Anything where the product's effect is physical but invisible.
The variable: The weather, and its host object

The arc

Beat	Name	What happens
0-3s	The normal room	A wide-ish handheld shot establishes an ordinary interior. The host object is present but not centred.
3-6s	First sign	The weather appears only at the object's boundary — a rime of frost at the glass base, a first drop under the plant — small and local.
6-10s	The spill	It crosses onto other real materials and behaves differently on each: frost fingers along wood grain, slows on metal, blooms on fabric.
10-13s	The room concedes	The light in the room changes to match the weather, not the other way round. Reflections and shadows update.
13-15s	Containment	The weather retreats to the object and stops at its edge, leaving one trace behind — a wet ring, a bent leaf.

Six values for the variable

frost creeping out of a glass across a table
a rain cloud tethered above a houseplant
heat haze pouring off a radiator down a corridor
fog pooling inside an open freezer drawer
a small wind that only moves one shelf
sunlight that only lands on one chair

Texture discipline

Physical, not stylised: real condensation behaviour, surface tension, the way frost follows grain. This is the one concept in the list that wants photographic realism rather than a drawn register.

The anchor

The light responds. This is the hard part and the thing to prompt explicitly — a microclimate that does not change the light reads as a sticker.

Negatives

no particle-system sparkle
no lens flares
no colour grading shift
no cuts
no music

It makes an invisible product effect visible without a diagram or a voiceover, and the escalation is bounded by the object itself, which is a natural end.

What to watch for

Weather is the effect models most want to make pretty. Every prompt beat should name a real material the weather is interacting with, or it drifts into wallpaper.

The sound becomes visible

Something audible in the room emits a form you can see, and the picture follows the audio.

Use it for: Audio products, acoustics, anything where the subject is a sound. Also the best of these for practising audio-led prompting.
The variable: The sound source

The arc

Beat	Name	What happens
0-3s	Sound only	The room is filmed plainly and the sound is established before anything visual happens. Nothing is drawn yet.
3-6s	First form	The form appears at the source and matches the sound's actual rhythm — a hum is continuous, a drip is discrete, a doorbell is two events.
6-10s	It travels	The form leaves the source and moves through the room along a path, changing shape as the sound reflects off hard and soft surfaces.
10-13s	Interference	A second sound enters and the two forms interact — cancelling, stacking, deflecting — without either becoming decorative.
13-15s	Silence	The sound stops. The form does not fade; it stops too, and the room is just a room again.

Six values for the variable

a fridge hum as a slow pulsing lattice
a dripping tap as a bouncing figure
a doorbell as a shape that runs down a hall
a bus passing as a wave through a window frame
a kettle rising as a column that fills the ceiling
a clock as a walking line

Texture discipline

Drawn but systematic: consistent line weight, a shape language that stays constant per sound type, no glow, no neon. Nothing that looks like an audio visualiser plug-in.

The anchor

Write the soundscape first, then write the picture to match it. This is the only concept here where the audio field is the primary document.

Negatives

no music of any kind
no waveform or spectrum graphics
no neon
no glow
no text

It inverts the usual order of work, which is instructive on its own. And because the sound is real and diegetic, the sync gives the model a hard timing reference for every beat.

What to watch for

Every model's instinct here is an audio visualiser. The negatives have to say so explicitly, more than once.

One shot, four decades

A locked-off frame in which the room ages around one object that never leaves.

Use it for: Longevity, heritage, repairability, anything whose claim is that it lasts.
The variable: The room and the surviving object

The arc

Beat	Name	What happens
0-3s	Now	A static, tripod-locked frame of the room as it is today. Deliberately unremarkable.
3-6s	Back one step	Surfaces, fittings and colours change to an earlier period around the object, in one continuous change rather than a cut.
6-10s	Back two	Further back. The light source changes character — daylight to tungsten to gaslight — and the room's use changes with it.
10-13s	The earliest	The oldest state, briefly, with the object visibly newer than everything else in frame.
13-15s	Return	Straight back to now in one continuous change, ending on the object, which is the only thing that has not moved.

Six values for the variable

a kitchen and a cast-iron pan
a workshop and a hand plane
a hallway and a mirror
a bedroom and a wooden toy
a shopfront and a brass handle
a garden and a stone step

Texture discipline

Each period carries its own capture register: colour response, grain, contrast and lens character consistent with the era, changing together rather than one at a time.

The anchor

The camera does not move at all. This is the only concept in the list that abandons the handheld conceit, which makes it a useful control: if a shot only works because of camera personality, this is where you find out.

Negatives

no camera movement
no cuts
no dissolves
no sepia wash as shorthand for the past
no captions or dates
no music

A locked frame gives the model a fixed geometry, which is the easiest condition under which to change everything else. It is technically the most achievable of these ten.

What to watch for

Models signal the past with a colour wash. Name the period through objects and fittings and let the capture register carry the rest.

The map redraws itself

A printed map animates its own route across the real surface it is lying on.

Use it for: Journeys, logistics, delivery, travel, migration, any process with a path.
The variable: The kind of document, and therefore the kind of journey

The arc

Beat	Name	What happens
0-3s	The paper	The map lies on a real surface with real creases and wear. A hand rests on it. Wind or light moves the paper slightly.
3-6s	The route lifts	One line detaches and travels off the paper onto the surface beneath, keeping cartographic weight and colour.
6-10s	The terrain arrives	Contours, hatching and symbols rise into low relief on the real surface — a hill becomes an actual raised ridge of ink.
10-13s	The traveller	A small symbol — a vehicle, a boot, a boat glyph — moves the route in real time, and the terrain reacts to it.
13-15s	Refold	Everything drains back into the printed map, which folds itself along its existing creases and lies flat.

Six values for the variable

a folded road atlas on a car bonnet
a hand-drawn walking route on a napkin
a transit map on a station wall
a nautical chart on a wet deck
a hiking contour map on a rock
a floor plan on a construction table

Texture discipline

Cartographic vocabulary: contour lines, hatching, stipple, grid ticks, a legend's line weights, spot colours of a printed map. Ink sitting on paper fibre, not glowing over it.

The anchor

The map's real creases and folds stay in shot and constrain everything — the route bends where the paper bends.

Negatives

no glowing GPS trails
no satellite imagery
no drop pins
no UI elements
no place names invented
no music

Cartography is a dense, specific visual language that almost nothing generated uses, and the fold-back gives a clean ending. The creases also give the model a physical logic to obey.

What to watch for

Generated maps produce fake place names and unreadable text. The negatives should exclude lettering entirely and let symbols carry it.

Scale collapse

Something tiny becomes something vast without a cut, on a matched texture.

Use it for: Scale claims, ingredients, materials science, anything where the small thing is the point.
The variable: The small thing, and the vast thing it matches

The arc

Beat	Name	What happens
0-3s	The macro	Extreme close-up on the small thing, shallow focus, with enough real context at the edge of frame to know what it is.
3-6s	The match	The camera pulls back and the texture holds — the same pattern reads at the next scale up, so the eye cannot find the seam.
6-10s	Two more steps	Two further pull-backs, each one reinterpreting the same pattern as something larger, without the pattern ever changing shape.
10-13s	The reveal	The widest scale is legible for the first time. Nothing has cut.
13-15s	Return	One continuous push back in to the macro, ending exactly where it started, with the original real context back at frame edge.

Six values for the variable

dust motes and a star field
a puddle and an ocean
sugar grains and dunes
a cracked glaze and a dry riverbed
fabric weave and ploughed fields
bubbles in dough and a cave system

Texture discipline

One texture throughout, described once and never varied. The only thing changing is what the texture is claimed to be.

The anchor

No cuts, and the pull-back never stops moving. The whole effect is continuity of pattern across a continuous move.

Negatives

no cuts
no dissolves
no cross-fades
no zoom blur
no space-documentary lighting
no music

It is the purest test of whether a model can hold a texture while changing everything about its context, and the round trip gives it a structure rather than just an effect.

What to watch for

The hardest of the ten. Models lose the texture at the second transition. Consider generating each scale step separately and anchoring on last-frame continuity rather than asking for one take.

Handwriting outlives the writer

Old handwriting lifts off its surface and re-enacts what it describes.

Use it for: Archives, letters, recipes, family history, anything where the document is the evidence.
The variable: The written surface

The arc

Beat	Name	What happens
0-3s	The document	Close, handheld, raking light so the paper fibre and the pen's indentation are visible. A hand turns it slightly.
3-6s	The stroke moves	One word's ink lifts and redraws itself in the air, in the writer's own hand, at writing speed rather than animation speed.
6-10s	It acts out	The handwriting forms the thing it describes — a place, a person, an event — still made entirely of the same ink line.
10-13s	The hand that wrote it	A drawn hand and sleeve form briefly from the same line, holding a pen, and then dissolve back into writing.
13-15s	Settling	The ink returns to the page and sits back into its own indentation, exactly aligned with where it was.

Six values for the variable

a letter in a shoebox
margin notes in a textbook
signatures on a plaster cast
a recipe in a family notebook
a tally scratched into a doorframe
graffiti under a bridge

Texture discipline

Ink on fibre: nib pressure variation, bleed into the paper, a dry patch where the pen ran out, uneven baseline, the specific wobble of an older hand.

The anchor

The handwriting must stay legible as handwriting throughout. If it becomes a shape, the idea is gone.

Negatives

no typed text
no new words invented
no calligraphy flourish
no parchment cliché
no candlelight
no music

Writing speed rather than animation speed is a small, unusual instruction that does most of the work, and the return-to-indentation ending is precise enough to be checkable.

What to watch for

Models cannot generate consistent legible text. Keep the visible words to a very short phrase, or frame so that only a few words are ever readable.

The shadow acts on its own

A real shadow stops matching its object, does something, and returns before it is caught.

Use it for: Character, mischief, imagination, anything with a playful register. The most comedic of the ten.
The variable: The light source that casts it

The arc

Beat	Name	What happens
0-3s	Correct shadow	A static-ish handheld shot in which object and shadow agree completely. The light source is visible or clearly implied.
3-6s	The lag	The object moves and the shadow follows a beat late, then over-extends the movement slightly. Nothing else is wrong yet.
6-10s	Independence	The shadow leaves the object entirely and does something the object is not doing, moving across real surfaces and bending correctly over their edges.
10-13s	Interaction	It interacts with another real shadow in frame, and the two behave as if they have mass.
13-15s	Snap back	The light shifts slightly and the shadow is instantly back where it should be, correct, as if nothing happened.

Six values for the variable

a desk lamp at night
low winter sun through blinds
a streetlight on a wet pavement
headlights sweeping a wall
a fire in a grate
a phone torch on a ceiling

Texture discipline

Pure silhouette. No interior detail, no gradient, no glow. Paper-cut edges with the softness the real light would actually give at that distance.

The anchor

The shadow's softness, direction and length must stay consistent with the visible light source at every moment, including while it misbehaves.

Negatives

no interior detail in the shadow
no glowing edges
no second light source appearing
no face on the shadow
no music

Silhouette is the cheapest register to keep consistent and the easiest to check — you can see instantly whether the model has broken the light. The lag in beat two is the same trick as the camera reacting late, applied to the subject.

What to watch for

Models add detail inside shadows. State that it is a flat silhouette in every beat, not just once at the end.

The recipe assembles itself

Ingredients arrive in order, in stop-motion, and the dish builds itself in one frame.

Use it for: Food, kits, anything with components and an order of operations.
The variable: The dish, and therefore the order

The arc

Beat	Name	What happens
0-3s	Empty surface	A real board or plate on a real counter, side-lit, with crumbs and wear visible. Nothing on it.
3-6s	Arrival	Ingredients walk, roll or hop into frame in stop-motion — held frames, small position jumps, no smooth interpolation — and queue.
6-10s	Build	They assemble in the correct order, each waiting its turn, with real weight and real settling when they land.
10-13s	The mistake and the fix	One ingredient arrives out of order, is nudged out by the next, and rejoins in the right place.
13-15s	Finished	The dish is complete and still. One held frame, then a hand enters and takes it.

Six values for the variable

a layered cake
a bowl of noodles
a sandwich in cross-section
a cocktail built in a clear glass
a lunchbox packed for a child
a plated dish deconstructed and rebuilt

Texture discipline

Stop-motion, stated as mechanics rather than as a style: held frames, a slight jitter between positions, tiny inconsistencies in placement, visible fingerprints in soft ingredients, no motion blur at all.

The anchor

No motion blur is the load-bearing instruction. It is what separates stop-motion from smooth animation, and it is the first thing a video model adds back.

Negatives

no motion blur
no smooth easing
no floating
no perfect symmetry
no advertising gloss
no music

Stop-motion is a register defined by its imperfections, which is exactly the kind of thing you have to ask for explicitly, and the wrong-order beat gives it narrative rather than just charm.

What to watch for

Every video model interpolates. Expect to fight for the held frames, and consider a lower frame rate at generation time rather than only asking in the prompt.

What each one contains, and why

A five-beat arc with timings, because five beats of roughly three seconds is the practical ceiling before a model starts silently dropping one.
The one variable, with six values, so the concept is a series rather than a film.
A texture discipline: the specific language that stops the model reverting to smooth advertising CGI.
An anchor: the single constraint that must hold or the idea collapses.
The negatives, and what to watch for, including where the concept is likely to be hardest.

The set that was actually run · Cinematic language · Video prompt craft