The Director's Worksheet Pack: 10 Free Worksheets for Cinematic AI Video
Ten fill-in worksheets that take you from wishlist prompter to AI video director — beat builder, camera decision sheet, carryover checklist, prompt compression triage, model routing plan, and the producer scorecard. Free, for Seedance, Runway and Dreamina.
By Ezekiel 'Kiel' Orji · 2026-08-02 · 16 min read
The Director's Worksheet Pack: 10 Free Worksheets for Cinematic AI Video
Most people using AI video are wishing, not directing. These ten worksheets are the difference. Each one is a lab from our live course Generative Cinematography — Directing, Not Prompting, and each is designed to be filled in for your own piece, not a class example. By the end, your completed sheets stack into a single shippable prompt — that is by design.
Every copy-paste template here also lives in the PxLabs prompt library under Video directing, one tap from copied.
Key takeaways
- AI video is a directing problem, not a generation problem: the model executes exactly what you specify, and every choice you don't make gets made for you, badly.
- One dominant motion per beat. The camera is specified five ways or it isn't directed. Emotion goes in as physical evidence, never as an adjective.
- Shots exist in a chain: environmental carryover is the largest single gap between amateur and professional AI prompting.
- Compression is a quality pass, and there is a short list of things you never cut.
- Ship at 95+ on the producer scorecard — until you are happy, not until you are tired.
How the pack works: everything in [SQUARE BRACKETS] is yours to fill. Fill them by hand while you are learning — the thinking is the training. Once the thinking is habit, hand a whole worksheet to your LLM with your project notes (the fill prompt is Worksheet 10's appendix). Never skip the review: the worksheets train your taste, and taste is the durable asset.
Worksheet 0 — The Director's Brief
Complete before writing any prompt, for every piece, forever.
A director who cannot answer these questions in one sentence each is not ready to prompt. This sheet takes five minutes and saves fifty generations.
0.1 — The Director's Question (the only question that matters)
What does the audience need to FEEL at the end of these [15] seconds?
One sentence. If you can't write it, stop here.
0.2 — The format spec (this becomes line one of your prompt)
[TITLE] — [DURATION]s [DELIVERABLE TYPE: trailer / music video / living
portrait / product cinematic], [ASPECT RATIO], [FRAME RATE if it matters].
0.3 — The closing thesis line (this becomes the LAST line of your prompt)
One sentence that names what the piece is FOR. It resolves the model's ambiguity, and it reminds you why you started when you're 47 trim iterations deep. Reference examples: "The growling IS the machine." · "The piece does not end. It becomes the painting." · "The city does not understand what hit it."
0.4 — The baseline generation. Write your piece as a naive wishlist prompt — the way you would have written it before any of this. Generate it. Label it BASELINE. Do not fix it yet. The distance between BASELINE and your final piece is the whole method.
✅ Done when: 0.1 and 0.3 each fit in one honest sentence, and BASELINE exists.
Worksheet 1 — The Failure Diagnosis Sheet
Use on every failed or disappointing generation, forever.
A failed generation is a diagnosis, not bad luck. Find your symptom; it names your fix.
| Symptom in the output | Probable cause | The fix |
|---|---|---|
| Motion is blurry mush / half-actions | Overloaded beat — more than one dominant action | Worksheet 2: rebuild under the hierarchy |
| Nothing happens / static drift | Verbs too abstract; no named motion | Name ONE physical motion with a velocity ("turns slowly") |
| Emotion missing or generic face | You prompted the emotion, not the evidence | Translate to physical evidence (below) |
| Face/outfit changes mid-shot | Identity drift — no lock, or over-density | Add identity block + "Identity locked, zero drift." Reduce density |
| Extra limbs, morphing objects | Hallucination trigger — ambiguity or contradiction | Find the two conflicting instructions; delete one |
| Looks like AI: too perfect, too smooth | No micro-imperfections written in | Worksheet 4, imperfection pass |
| Crowded, protagonist lost | Density failure | One dominant subject + one supporting detail, rest to negative space |
| Style flickers between looks | Over-stylization / competing registers | "One unified register only" + cut the second style |
| Beats feel like unrelated clips | No environmental chaining | Worksheet 4: write the carryover sentences |
The evidence translation drill. The emotion never goes in the prompt — the physical evidence does:
Grief → Eyes lower. Shoulders drop one centimeter. A tear catches light at the corner of her eye but does not fall. Her hand rises halfway to her face and stops.
Fill for your piece: [YOUR EMOTION] → [3–4 pieces of physical evidence: posture, breath, gaze, a gesture that stops].
✅ Done when: you can look at any bad generation and name its row without guessing.
Worksheet 2 — The Beat Builder
One sheet per beat. A 5-beat piece = 5 of these.
Every shot gets one dominant motion and at most two supporting motions. More, and the model averages into mush.
BEAT: [ALL CAPS MEMORY ANCHOR — e.g., THE DECIPHER] [0:0X–0:0Y]
1. CAMERA MOTION (dominant — the audience's perceptual experience):
[One move, one sentence. If you cannot name it in one sentence,
you have not finished thinking.]
2. CHARACTER GESTURE (the single gesture that earns the beat):
[One action. A turn. A blink. A hand rising. ONE.]
3. SECONDARY MOTION (jewelry / fabric — the life signal):
[e.g., earrings swing once; the robe drifts]
4. ENVIRONMENTAL MOVEMENT (atmospheric breath, never competing):
[e.g., dust drifting in light shafts]
5. VFX ACCENT (lowest priority — or NOTHING):
[flare / particle / light shift]
The "and" test: read your beat aloud. If "and" connects two ACTIONS, the beat is overloaded — demote one to a parenthetical life signal or move it to the next beat.
Assembled beat syntax:
[0:0X–0:0Y] [BEAT NAME]. [Character gesture as the sentence].
(Secondary: [secondary motion]. Background: [environmental movement].)
[Camera motion]. [Light for this beat]. Sound: [sound implication].
✅ Done when: every beat passes the "and" test and the camera is line one of every sheet.
Worksheet 3 — The Camera Decision Sheet
The camera is the altar. Choose the movement that emotionally narrates the beat — then specify it fully.
The emotional matching table:
| If the beat needs to say… | The camera says it with… |
|---|---|
| "Witness this. It is bigger than you." | Locked tripod (sacred witness) |
| "Come closer. Something is worth seeing." | Slow forward push (emotional gravity) |
| "Step back. The scale is more than you knew." | Slow pullback (revelation) |
| "This being is mythic. Circle them." | Orbital drift (mythic perception) |
| "This is real. A body held this camera." | Handheld (presence) |
| "Look at THIS. The detail is the story." | Macro insert |
| "Something is wrong, or about to be." | Dutch tilt |
| "Your attention belongs elsewhere now." | Rack focus |
| "Sudden importance. No warning." | Crash zoom (use rarely) |
The full specification — a move is not directed until all five are named:
Move: [from the table]
Lens: [e.g., 35mm anamorphic / 100mm macro]
Framing: [eye-level / low angle / one-point / asymmetric, subject frame-left]
Velocity: [slow / gently closing distance / across full duration]
Breath: [subtle anamorphic breathing / slight handheld / locked, no breath]
The "when NOT" check: does your chosen move fight the beat's emotion? A whip pan in a grief scene. A crash zoom in a sacred moment. If yes, return to the table.
✅ Done when: all five lines are filled and the move matches the feeling from Worksheet 0.1.
Worksheet 4 — The Carryover Checklist
Shots do not exist in isolation; they exist in a chain. This is the largest single difference between amateur and professional AI prompting.
The carryover grid. For each pair of consecutive beats, name what carries over — light, wind, dust, fabric momentum, direction of movement, emotional state. "Nothing" is not an answer; something always carries.
The carryover sentences. Turn each row into one sentence at the start of the receiving beat:
"The dust from the collapse still drifts through the light shafts as she…"
"The warm firelight from the previous frame holds; her scarf is still settling from the turn."
The imperfection pass. The model defaults to perfection, and perfection reads as AI. Choose 2–3 (restraint is also discipline): slight handheld breathing · anamorphic lens breathing · rain-flecks or dust on the lens glass · missed focus briefly catching an edge object · slightly rotated furniture, compressed cushions · imperfect timing, the gesture landing a half-beat late.
The density audit. Per beat: one dominant visual subject ✅ / at most one supporting detail ✅ / everything else surrendered to negative space, aim 60%+ ✅. The audience doesn't feel deprived by sparseness; they feel attention.
✅ Done when: every grid row has a named carryover, the sentences are written into the beats, and every beat passes the density audit.
Worksheet 5 — The Assembly & Compression Sheet
The worksheet that turns all previous sheets into a shippable prompt.
Assembly order — the model weighs the beginning heavily and reads the end last:
- Title & format spec ← Worksheet 0
- Style & mood register — "Hollywood-tier sci-fi epic. Film grain, lifted blacks, warm halation. Photographic, not CGI."
- Identity block(s) —
[SUBJECT]: [locked description — age, skin, hair, wardrobe, one signature detail]. Identity locked, zero drift. - Audio architecture (if relevant)
- Color hierarchy — percentages: 40% black system, 35% brown void, 20% gold, 5% accent. Color is hierarchy, not decoration.
- Global camera instruction ← Worksheet 3
- Beat-by-beat timed blocks ← Worksheets 2 + 4, carryovers in place
- Lighting summary
- CONSTRAINTS block — the last word the model reads: the power position
- Closing thesis line ← Worksheet 0.3
Formatting discipline: ALL CAPS block headers, [bracketed] timestamps, label: value pairs — functional, not aesthetic; the model chunks the prompt by these markers.
Beat timing scaffold (15s trailer): 0:00–0:02 HOOK · 0:02–0:05 INCITING · 0:05–0:09 DEVELOPMENT · 0:09–0:12 STAKES · 0:12–0:15 LANDING — always ending on a title card or thesis image, never an unanchored fade.
The compression triage. Validate the character count (targets: Seedance/Dreamina 3950–3999, Runway 3450–3499). If over cap, cut in this order and only this order: decoration words → redundant adjectives → semantic compression (conceptual words become visual words) → the weakest beat loses its secondary motion → a whole beat.
Never cut, under any circumstances: identity locks · camera specifications · the constraints block · the closing thesis line · carryover sentences.
✅ Done when: the prompt is in the 10-part order, under cap, and everything on the never-cut list survived.
Worksheet 6 — The Routing Plan
For any piece bigger than one model can carry. Production at scale is not a prompting problem — it is a routing problem.
| Need | Route to | Why |
|---|---|---|
| Motion-heavy environment / atmosphere | Seedance-class video model | Sustained atmospheric continuity, cinematic camera grammar |
| UI / HUD overlays, readable text | Interface-strong image model (keyframes extracted, overlaid) | Cleaner graphic consistency than video models |
| Character close-ups across many shots | Strongest identity-preservation model | Facial consistency is the scarcest resource — spend your best tool on it |
| Title cards, branded styling | Design/image tool | Type should never be left to a video model |
Then fill the shot routing table for your piece: each shot, what it demands most, where it routes, whether it needs an identity lock, and whether a keyframe hands off in or out.
The translation note: when one prompt travels across platforms, the structure and hierarchy stay identical; only the cap, the vocabulary, and platform constraint syntax change. The directing does not change. Only the dialect does.
✅ Done when: every shot has a routed home and a reason.
Worksheet 7 — The Personal Method Document
The sheet you keep for years. Start it now, never finish it.
A method = repeatable disciplines + a protected phrase library. Draft yours:
- My materials — the 5–8 surfaces and textures my world is made of
- My color hierarchy — with percentages; color is hierarchy, not decoration
- My camera posture — observer? witness? intimate? mythic orbit?
- My negative space — what my emptiness is FOR
- My reference stack — 5 films / photographers / painters my frames answer to
- My protected phrase library — verbatim phrases that consistently produce results in MY work. Seed examples: "Identity locked, zero drift." · "The camera is the altar." · "60%+ negative space per frame as psychological memory-space." · "Imperfect, handmade, lived-in."
- My single defining sentence — the method in one line
✅ Done when: all seven have a first honest answer. They will all change. That is the point.
Worksheet 8 — The Producer Scorecard
Run on EVERY generation before it ships. No exceptions.
Rate /100 per category. Below 80 in any category = that category is your revision note. Ship-grade is 95+ across all seven.
| # | Category | The question |
|---|---|---|
| 1 | Composition | Does the framing earn the beat? |
| 2 | Motion | Is camera + character motion specific, not averaged mush? |
| 3 | Light | Are source, temperature, quality AND behaviour specified and visible? |
| 4 | Material | Are the surfaces specific to this world, not generic AI sheen? |
| 5 | Continuity | Does the beat chain to its neighbours — carryover visible? |
| 6 | Emotion | Does the physical evidence carry the intended feeling? |
| 7 | Discipline | Negative space respected? Hierarchy clean? One register? |
Scoring honesty rules: score the GENERATION, not your intention. Score before you get attached. A 79 is a 79 — rounding up is how average work ships.
✅ Done when: you can defend every number out loud in one sentence.
Worksheet 9 — The Ship Sheet
The final assembly. One page that proves the whole method.
- The Director's Question answered in one sentence
- Thesis line written and sitting as the prompt's final line
- Every beat built on a Beat Builder sheet; every beat passes the "and" test
- Camera specified five ways and emotionally matched
- Carryover grid complete; imperfections chosen; density audit passed
- Assembled in the 10-part order; constraints block in the final position
- Under cap, nothing from the never-cut list sacrificed
- Routing plan (if multi-model) executed with identity locks at every handoff
- Producer scorecard: 95+ in all seven categories — until you are happy, not until you are tired
Worksheet 10 — The LLM-Assisted Fill
Once a worksheet's thinking is second nature, populate it faster by handing the sheet plus your project context to your LLM:
Here is a worksheet from my AI video directing course, and here is my
project: [PASTE YOUR DIRECTOR'S BRIEF — Worksheet 0 — plus any notes,
references, or completed earlier worksheets].
Fill in the bracketed fields of this worksheet for MY piece. Where the
worksheet demands a choice (a camera move, a dominant action, phrases
to protect), propose your best option AND one alternative, with one
sentence of reasoning each, so I can direct the final call.
WORKSHEET:
[paste the full worksheet]
Then review every field. The LLM drafts; you direct. If you find yourself approving fills you don't understand, go back to filling that worksheet by hand — the shortcut has gotten ahead of the skill.
Where to go from here
These worksheets are the lab materials for Generative Cinematography — Directing, Not Prompting, our live course where Kiel walks a cohort from a naive BASELINE to a shipped, producer-rated piece across nine sessions. If you want the full method — the model behaviour analysis, the case studies, the filter-safe brand workflows — that's where it's taught, live, on your own piece.
The companion reading is free: Director's Prompt: The AI Secret Sauce, the complete manual this pack is built on, and the free lesson that summarises it.
And the room these methods get sharpened in is the PxLabs community — the builders' forum, the project showcase, every template in this post in the copy-paste prompt library, Thursday Vibe Sessions and meet-ups across Nigeria. ₦30,000/year.
The camera is the altar. The world is the star. The work is the work. Ship.