AI Video Prompt Templates: Copy-Paste Library and Method

Copy-paste AI video prompt templates, the keyword clusters that control light and motion, and the Afrofuturist method behind cinematic AI ads from Lagos.

By Ezekiel 'Kiel' Orji — Co-founder, PxLabs · 2026-07-29 · 10 min read

AI Video Prompt Templates: Copy-Paste Library and Method

Most AI video prompts fail for the same reason: they are written as wishes, not as directions. You type "cinematic shot of a woman in a beautiful futuristic city" and hope the model reads your mind. It never does. What works instead is a system — reusable prompt templates with locked structure, a keyword library that controls the camera, the light and the motion, and a personal method that gives the work a point of view no model can generate on its own.

This is Part 3 of our Director's Prompt: The AI Secret Sauce series at PxLabs, and it is the part you can put to work today. Whether you are a solo creator who wants copy-paste starting points, a marketing team building an internal prompt library, or someone automating content production at scale, everything below is designed to be lifted straight into your workflow. It comes from years of shipped work — trailers, fashion films, branded packshots — made from Lagos, for African and global audiences.

Key takeaways

The Afrofuturist method: cultural specificity is the spine

Before the templates, the method — because a template filled with generic content produces generic video. This is the DNA that runs through everything we make and teach.

The core belief: Afrofuturism is not aesthetic decoration. It is a cinematic language. The visual vocabulary of post-colonial Black futurism has specific compositional rules, colour hierarchies, material languages and emotional registers that have to be honoured. When done right, the work feels mythic and intimate simultaneously. The Yoruba woman in the brown void is a goddess. The astronaut whose holographic face glows inside her visor is humanity preserved.

Cultural specificity as the spine: a veiled ultramarine figure offers a luminous tablet to workers below, ringed in amber halos — mythic and intimate at once, nothing generic about it

The material language of the future-African

The future-African does not look like Western sci-fi. It looks like a future that grew from African material traditions: carved, not printed; forged, not assembled; sacred, not commercial. In practice that means prompting with a specific material palette — terracotta and volcanic stone, obsidian flooring, programmable gold filigree threading through clay seams, aso ebi cloth as sisterhood made visible, hand-thrown bronze with the maker's hand still in it, smoked glass and brushed bronze so the futuristic surfaces stay warm, never cold.

Tactility is specified, not implied. Name materials, name imperfections, name the maker's hand: "hand-thrown bronze vessels," "brushstroke texture visible on the gold leaf," "imperfect, handmade, lived-in." These instructions tell the model that nothing in this world was mass-produced. Everything has weight.

Colour as hierarchy, not decoration

The colour science, codified as compositional law:

The percentages express compositional discipline, and the qualitative locks ("NOT flat black," "NOT modern luxury gold") prevent specific AI failure modes. When you specify colour as hierarchy rather than decoration, the model produces images that feel composed — not arranged, composed.

Negative space, and the camera as witness

Two more rules complete the method. First, 60%+ negative space per frame. Brown emptiness is not background; it is psychological memory-space. Most prompters try to fill the frame; the discipline is to leave it mostly empty and let one subject anchor it — the result reads as devotional. Second, the camera is a witness, not a participant: locked tripod, slow push, slow pullback, orbital arc. No whip pans, no crash zooms. And when the surreal appears, it is grounded: if one element in a beat is surreal, every other element should be real — real materials, real light, real physics.

The whole method compresses into one sentence: "Camera is the altar. World is the star. Every streak is a memory." Hold those three principles as you write and you will write Afrofuturist cinematography. Abandon them and you will write generic AI video.

What real projects taught us (case-study lessons)

Every rule above was earned on shipped work. A tight selection of the lessons:

▶ Type II — the motion experiment behind that lesson: planetary-scale structure kept human by one intimate beat

The keyword clusters that control light, lens and motion

You do not need a thousand keywords. You need eight clusters, used deliberately. These are the strongest, with when to reach for each:

  1. Camera and lens — use when a beat feels visually flat. Assign a motivated lens per beat: "14mm ultra-wide," "27mm cathedral wide," "50mm portrait," "100mm macro"; moves like "locked tripod," "slow forward push," "orbital arc," "rack focus."
  2. Lighting sources and qualities — use when renders look like flat CGI. "Single key from upper-left," "volumetric god-rays through architectural gaps," "warm molten-gold firelight from hidden practicals off-frame," "Rembrandt lighting," "single key, no fill," "museum-lighting logic."
  3. Texture and material — use when the world feels mass-produced. "Terracotta composite," "oxidized bronze," "obsidian flooring," "gold leaf accents," "visible canvas texture," plus imperfections: "imperfect, handmade, lived-in," "compressed cushions," "lifted blacks," "warm halation."
  4. Motion (camera, character, environmental) — use when clips feel static or chaotic. "Eyes find the lens at [time] and HOLD," "knowing breath, micro-smile barely registering," "fine sand drifts," "naira notes suspended mid-fall," "earrings swing once as head turns."
  5. Pacing and transitions — use to set tempo before the model invents its own. "Slow contemplative pacing," "held three seconds," "one unbroken camera breath, no cuts" for reverence; "frame jerks," "camera makes mistakes," "cut on motion" for urgency.
  6. Realism register — use to stop style drift between photoreal, painterly and found-footage. "Photographic, not CGI," "cinema look, warm film grain, halation, lifted blacks," "one unified painterly register only," "handheld, raw, no stabilization."
  7. Cinematic mood — use to set the emotional register as global state at the top of the prompt. "Sacred social sanctuary that already exists," "reverent observation," "documentary stellar physics," "recovered visual memory from another era."
  8. Continuity and global state — use on any multi-beat piece. "Carries forward: [element]," "same key light across all beats," "wind direction frame-right to frame-left throughout," plus global blocks: LIGHTING, COLOR HIERARCHY, ATMOSPHERE, CAMERA, CONSTRAINTS.

Clusters doing their separate jobs in one frame: a boy amid floating crystalline debris and giant broken clock faces — material, motion and mood each specified, none left to chance

Marketing teams: these eight clusters are your internal prompt library's table of contents. Put each in a shared doc, add your brand's specifics under each heading, and every prompt anyone on the team writes starts from the same vocabulary.

A master prompt template you can copy today

This is the fashion film template, verbatim from the working method — the same architecture behind the OWAMBE piece. Drop your specifics into the brackets:

[TITLE] — [DURATION]s painterly fashion film, [ASPECT RATIO]. Single continuous shot, slow forward push. [NUMBER OF BEATS] sustained held moments. Hypnotic, observational. Recovered visual memory.

THE SUBJECT: [identity block]. [Skin/feature treatment with "NOT" specifications]. [Wardrobe identity locked]. [Posture or gestural defaults]. Identity locked.

[OTHER FIGURES IF ANY]: [Description with same painterly register]. All in same [style/cloth/world]. Distinct individuals, not clones.

STYLE: [primary aesthetic — e.g., "Hyperreal painterly portraiture, 2D hand-painted, visible-but-controlled brushwork on background, smoother on skin"]. [Most refined treatment on subject's face — focal painted object]. No CGI gloss. No photorealism. One unified register.

COLOR HIERARCHY: [percentages and qualitative locks]. Color is hierarchy, not decoration.

CAMERA: Single continuous slow forward push across [duration]. Subtle anamorphic breathing. No cuts.

[BEAT 1 — PRESENCE] [held moment of subject alone in negative space]

[BEAT 2 — DEVELOPMENT] [environmental detail or supporting figures enter at frame edges]

[BEAT 3 — RESOLUTION] [either an emotional landing or a snapshot freeze]

CONSTRAINTS: One register only. Color as hierarchy. [60%+ negative space per frame]. Identity locked. No CGI gloss. No saturation drift.

▶ Nextberries Fashion Film — exactly this register of prompt in action: the painterly fashion-film architecture above, filled in and shipped

Anatomy: what each clause does

For teams making product ads, the same architecture adapts directly: the product cinematic variant swaps the subject block for @image_[N] = PRODUCT. [Generic description of the product without brand name]. Match exactly. Identity locked, zero drift. and ends its packshot beat with "Label readable, never deforming" — the Captain Morgan filter-safe pattern built straight into the template.

Automating content creation: from single prompts to scene systems

If you are automating content production — batch-generating campaign variants, localising one concept across markets — the unit of work is no longer the prompt. It is the scene system: a structured set of prompts that share character identity locks, world specifications, colour hierarchies, atmospheric continuities and camera grammars. You write the system once; each shot inherits from it. Build a world bible per project, reuse it verbatim across every prompt, and validate each generation against it before shipping.

Two production disciplines make this reliable at scale. First, validate length before you paste: target 3450–3499 characters for Runway and 3950–3999 for Seedance, and if you are over cap, cut decoration words, then redundant adjectives, then the weakest beat. Second, rate every generation across seven categories out of 100 — composition, motion, light, material, continuity, emotion, discipline. Anything under 80 in a category tells you exactly where to revise. Producer-grade work is 95+ across the board.

Where this is going: prompting becomes directing

The first generation of AI video was about prompting: write what you want and hope. The second generation is about directing: write what you want with such specificity that the model has no degrees of freedom. In five years, "prompt engineering" as a separate discipline will probably not exist. What will exist is the role that has existed since the silent era — the person who chooses what the audience sees.

The skills that will transfer: cinematographic vocabulary, motion hierarchy, colour discipline, continuity logic, emotional pacing, material specification, audience attention management. The skills that will become obsolete: prompt tricks, keyword stuffing, model-specific workarounds. Invest in the durable skills; treat the platform hacks as scaffolding.

Every limitation will eventually be solved — identity drift, continuity, long durations. When that happens, the bottleneck shifts entirely to taste. And the question that outlives every platform change: what does the audience need to feel at the end of this 15 seconds? If you can answer that in one sentence, the prompt writes itself.

Cinema is the art of making time feel like something. AI video is cinema, eventually — we are still in its silent era. You have inherited a working method. Now do better than we did with it.

Next in the series

Part 4 puts this method under commercial pressure: the brand-campaign case studies — how these templates and keyword clusters perform on real ads for real clients, from filter-safe packshots to heritage campaigns, with the numbers and the revisions shown. Read it next: How the method performs on real brand campaigns.

Learn this live

Reading a template is one thing; directing a model in the room with someone who ships this work is another. At PxLabs in Lagos we teach exactly this method, live:

Learn this live

PxLabs runs live AI courses in Lagos and online covering exactly this material, plus enterprise AI training for teams.

Previous: Seedance vs Runway vs Dreamina: Best AI Video Model for Ads

Next: How to Make AI Ads: 4 Real Brand Campaign Case Studies

All posts in the series


Open this page on PxLabs →