AI Video Prompt Formula That Actually Transfers Between Models: A "Slots + Test Grid" Method for Veo3Gen

TL;DR

A prompt transfers between video models when it’s structured and tested, not when it’s longer. Use a portable 6-slot prompt card (Subject, Action, Scene, Camera, Lighting, Style) based on FlexClip’s structure (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos), then run a 3×3 Test Grid (9 generations) that changes only Action and Camera. You’ll learn—empirically—what the model actually obeys, and you’ll be able to “translate” your intent to other tools by varying one slot at a time.

Key takeaways

Use a slot-based AI video prompt formula: Subject + Action + Scene + (Camera Movement + Lighting + Style) (FlexClip) (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos).
Treat Action as the core: FlexClip calls Action the core because it drives the storyline—so when motion is weak, rewrite Action first (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos).
Treat camera terms as hypotheses: FlexClip defines camera movement as shot/angle/movement and notes you can combine moves, but models interpret terms differently—verify with tests (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos).
For image-to-video, anchor to the image and write one line: “describe what changes” (FlexClip’s image-to-video structures) (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos).
Don’t “iterate randomly.” Log 9 quick runs, pick winners, and save them as a personal prompt spec you can reuse.

Why prompts fail to transfer between models

Most “prompt advice” is really a style of writing, not a method of learning.

What breaks across models is usually:

Action ambiguity (the clip looks static because nothing observable was requested)
Camera language mismatch (the model ignores or misreads jargon)
Constraint collisions (style drift because you asked for incompatible aesthetics)

FlexClip’s guidance is a useful baseline because it forces the essentials: Subject + Action + Scene + (Camera Movement + Lighting + Style) (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos). The fix is to:

turn that into a strict slot card you can reuse, and
add a tiny experiment: the 3×3 Test Grid.

The 6-slot prompt card (your transferable “prompt spec”)

Write each slot so a stranger could storyboard it. Keep each slot to 1–3 short lines.

Slot 1 — Subject (who/what is on screen)

Write a cast list + identifiers you can see.

Good Subject lines

“A bike courier in a neon rain jacket, soaked hair, delivery bag.”
“A matte-black wireless earbud case with a small front LED, on a pale oak desk.”

If identity drifts

Add 2–3 concrete identifiers (wardrobe/prop/material/color)
Remove vague adjectives that compete (e.g., “mysterious,” “cool,” “epic”)

Slot 2 — Action (the visible change)

FlexClip calls Action the core because it drives the storyline (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos). Translate that into an operator rule:

Action must contain a visible change the viewer can’t miss.

Three Action patterns that test well

Single change: “The lid flips open.”
Start → end: “Starts closed → ends fully open facing camera.”
One beat + outcome: “Snaps the paper map open, holding it toward camera.”

If motion is weak

Delete extra beats. Keep one change.
Replace “interacts with” with a verb you can see: “opens,” “turns,” “drops,” “steps,” “points.”

Slot 3 — Scene (where it happens)

Anchor place/time/materials.

Good Scene lines

“Narrow city alley at night, wet asphalt, neon reflections in puddles, light fog.”
“Bright home office, pale oak desk, minimal clutter, daylight from a left window.”

If the scene drifts

Move the key nouns earlier (“alley at night, wet asphalt…”) and cut decoration.

Slot 4 — Camera (framing + movement)

FlexClip defines camera movement as shot/angle/movement that adds narrative and visual appeal, and notes movements can be combined (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos). Use that—then verify it.

Camera rule: one framing + one movement.

Good Camera lines

“Locked medium shot, eye-level; subject centered.”
“Slow push-in from medium shot to medium close-up.”
“Slow orbit clockwise, keeping subject centered.”

If camera is ignored

Reduce to locked and make Action do the work.
Rewrite jargon into an observable outcome (example below).

Slot 5 — Lighting (physical, not poetic)

One key + one support is enough.

Good Lighting lines

“Neon key from camera-left; streetlight rim from behind.”
“Soft diffused daylight from left; gentle shadows; no harsh highlights.”

Slot 6 — Style (finish + constraints)

Pick one lane and add simple “no’s.”

Good Style lines

“Cinematic realism; restrained color grade; no surreal elements; no text overlays.”
“Clean commercial product video; minimal; crisp edges; no hands; no on-screen text.”

The 3×3 Test Grid: 9 runs that replace guessing with evidence

Your goal isn’t to find a magic prompt. It’s to learn what this model reliably obeys for this concept.

You’ll keep Subject / Scene / Lighting / Style identical for all 9 runs.

You will vary only:

Rows: Camera
Columns: Action specificity

Copy/paste Test Grid table

	Action A: minimal	Action B: visible change	Action C: start → end
Camera 1: locked	A1	B1	C1
Camera 2: push-in	A2	B2	C2
Camera 3: orbit/pan	A3	B3	C3

Action column examples

A (minimal): “He looks around.”
B (visible change): “He snaps a paper map open.”
C (start→end): “Starts with map folded → ends with map fully opened facing camera.”

Camera row examples

Camera 1: “Locked medium shot, eye-level.”
Camera 2: “Slow push-in from medium to medium close-up.”
Camera 3: “Slow orbit clockwise, keeping subject centered.”

What to log (fast, consistent)

For each cell, write three lines:

Action obeyed? Yes / Partial / No (what happened)
Camera obeyed? Yes / Partial / No (what happened)
What broke? (identity drift, scene drift, style drift, weird physics)

After 9 runs you’ll have a practical answer to:

“How explicit must Action be to get motion?”
“Which camera phrasing is actually honored?”
“Which slot triggers drift?”

If you want to scale this beyond manual testing, n8n shows an automation pattern that pulls ideas from Google Sheets and turns them into finished videos with captions/voiceovers/descriptions on a schedule (https://n8n.io/workflows/3442-fully-automated-ai-video-generation-and-multi-platform-publishing/). You’re copying the pattern: structured input → repeatable runs → logged output.

Worked example (before/after + test grid you can actually run)

This is the part most posts skip: a concrete conversion from “vibes” to a reusable spec.

Before (vibe prompt)

“Cinematic cyberpunk courier in neon alley, dramatic lighting, 35mm, dolly in, hyperreal, very detailed, cool vibes, rain, intense.”

What’s missing

No Action you can verify
Camera jargon without an observable outcome
Style pile-up (“hyperreal,” “very detailed,” “intense”) with no constraints

After (portable 6-slot prompt card)

Use this as your base card for all 9 grid cells.

Subject

A bike courier in a neon rain jacket, soaked hair, delivery bag.

Action (swap per grid column)

B version: He snaps a paper map open.

Scene

Narrow city alley at night, wet asphalt, neon reflections in puddles, light fog.

Camera (swap per grid row)

Camera 1 version: Locked medium shot, eye-level; subject centered.

Lighting

Neon key from camera-left; streetlight rim from behind; rain reflections on ground.

Style

Cinematic realism; restrained color grade; no surreal elements; no text overlays.

Fill the grid (exact substitutions)

Keep the card identical except the two lines below.

Camera substitutions

Row 1: “Locked medium shot, eye-level; subject centered.”
Row 2: “Slow push-in from medium to medium close-up.”
Row 3: “Slow orbit clockwise, keeping subject centered.”

Action substitutions

Col A: “He looks around.”
Col B: “He snaps a paper map open.”
Col C: “Starts with map folded → ends with map fully opened facing camera.”

Now you can say something meaningful like:

“For this concept, the model obeyed camera only when Action was minimal,” or
“Start→end phrasing fixed motion but broke identity unless camera was locked.”

Camera translation trick (when jargon doesn’t transfer)

FlexClip notes you can combine moves (e.g., “move down and zoom out”) (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos). If your model ignores “dolly zoom,” rewrite as:

Framing + subject position + change over time

“Start medium shot with subject centered → end close-up, subject still centered; background feels compressed.”

You’re no longer asking the model to know film-school terms; you’re describing what the viewer should see.

Text-to-video: 3 starter prompt cards (and what to test)

These are deliberately short so they fit the slots and the grid.

Card 1 — Product micro-demo (commercial)

Subject: Matte-black wireless earbud case with a small front LED on a pale oak desk.

Action: Lid flips open.

Scene: Bright home office, minimal clutter.

Camera: Top-down shot; locked.

Lighting: Soft daylight from left; controlled highlights.

Style: Clean commercial; crisp edges; no hands; no text overlays.

Test focus

Action: “lid flips open” vs “starts closed → ends open”
Camera: locked top-down vs subtle push-in

Card 2 — Talking-head moment (audio intent)

Subject: A founder in a simple studio setup, neutral wardrobe.

Action: Looks into camera and delivers one short line.

Scene: Plain background with subtle depth.

Camera: Medium close-up; locked.

Lighting: Soft key + gentle fill; catchlights.

Style: Natural documentary; no beauty retouch; no text overlays.

Card 3 — POV “how-to”

Subject: Two hands holding a foldable phone stand.

Action: Starts folded → ends locked open.

Scene: Kitchen counter, minimal clutter.

Camera: POV; hands centered; locked.

Lighting: Soft overhead practical.

Style: Realistic instructional; no overlays.

Image-to-video: “describe what changes” (FlexClip structures)

FlexClip’s image-to-video structures emphasize explicit subject/action/background/camera components (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos). In practice, your image already supplies the subject and scene—so your prompt should focus on the delta.

Image-to-video card 1 — Subtle “bring it alive”

Subject/Scene: Keep exactly as the input image.
Action (describe what changes): “Only subtle motion: subject blinks once; hair moves slightly in a breeze.”
Camera: “Locked.”
Style/Lighting: “Match the image; realistic.”

Image-to-video card 2 — Background movement + parallax

Subject/Scene: Keep exactly as the input image.
Action: “Background neon flickers; light rain falls; puddles ripple.”
Camera: “Very slow push-in; subtle parallax.”

Image-to-video card 3 — Explicit start→end

Action: “Start exactly as input image → end with subject turning head ~15 degrees toward camera, maintaining identity.”
Camera: “Locked.”

Where Veo3Gen fits (when you’re ready to run this method at scale)

If you’re doing lots of grid tests or variants, Veo3Gen is positioned as an affordable way to access Google’s Veo 3.1 models without Google’s enterprise pricing. It offers three modes—Veo 3.1 Fast, Veo 3.1 Quality, and Veo 3.1 Lite—and supports text-to-video and image-to-video, including first-and-last-frame control on Veo 3.1. Supported resolutions are 720p, 1080p, and 4K (4K on Fast/Quality) with 16:9 and 9:16 aspect ratios. Generations include native, synchronized audio (dialogue, SFX, music) in a single pass. There’s also a developer API for programmatic generation, and pricing is pay-as-you-go credits plus optional monthly plans with non-expiring purchased credits; new users get free credits to start.

CTA (mid-article): If you want to run the Test Grid without bolting on a separate audio step, try Veo3Gen for a few 9-run experiments and save your winning cards as a reusable prompt spec.

Checklist

Write the 6-slot card: Subject / Action / Scene / Camera / Lighting / Style
Make Action one visible change (or a start → end transition)
Keep Camera to one framing + one movement
Run the 3×3 Test Grid (9 generations) varying only Action + Camera
Log three lines per run: Action obeyed / Camera obeyed / What broke
Rewrite camera jargon into observable outcomes when needed
Convert winners into a one-page personal prompt spec

FAQ

How do I write an AI video prompt formula that works across models?

Use FlexClip’s structure—Subject + Action + Scene + (Camera Movement + Lighting + Style)—as strict slots, then test Action/Camera variations in a small grid so you learn what each model actually obeys (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos).

What’s the fastest way to get more motion in text-to-video?

Rewrite Action first. FlexClip calls Action the core because it drives the storyline—so make Action a single visible change or a start→end transition (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos).

Why do camera moves get ignored even when I’m specific?

Camera language is interpreted differently across models. FlexClip frames camera movement as shot/angle/movement and shows moves can be combined, but that doesn’t mean your model will honor your terms—verify with a grid and simplify to one move (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos).

How do I prompt image-to-video without changing the image?

Anchor subject/scene to “keep exactly as the image,” and put your intent into a single “describe what changes” Action line (blink, breeze, subtle parallax). FlexClip provides image-to-video prompt structures that emphasize explicit action/background/camera components (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos).

How can I automate daily prompt testing and publishing?

Use the same principles as the n8n workflow template: structured ideas in a sheet, repeatable generation steps, and automated publishing outputs. The template describes pulling ideas from Google Sheets on a schedule and producing publish-ready assets with captions and descriptions (https://n8n.io/workflows/3442-fully-automated-ai-video-generation-and-multi-platform-publishing/).

Do I need separate tools for audio?

Some workflows add audio later, but Veo3Gen generations can include native, synchronized audio (dialogue, SFX, music) in a single pass, which can simplify your iteration loop.

Closing: turn winning prompts into a library (not a one-off)

Run two or three Test Grids in your niche, then freeze what worked into a one-page prompt spec: your go-to slot order, your best Action patterns, and the camera phrasing your model actually obeys.

CTA (closing): If you want a practical setup for repeating that loop—testing variants in 16:9 or 9:16, generating up to 4K (Fast/Quality), and keeping audio synchronized in one generation—use Veo3Gen to build and batch-run your prompt cards, then reuse the winners across campaigns.

Start creating with Veo3Gen

Veo3Gen gives you affordable Veo 3.1 video generation with native audio, up to 4K, and credits that never expire — with free credits to start.

Generate your first video now: Get started
Compare plans and pay-as-you-go pricing: See pricing