Poe’s “Descriptive, Not Instructive” Video Prompt Rule—A Creator Rewrite Playbook for Veo3Gen (as of 2026-05-08)

When creators say “the model ignored my prompt,” what they often mean is: they wrote instructions the system can’t reliably “obey,” instead of a description it can confidently “render.” Poe’s Creator guidance for video prompting leans hard into this: describe the desired video rather than instruct the bot. (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)

This post is a practical rewrite playbook for Veo3Gen: you’ll take common command-style lines like “make it cinematic,” “add a dolly zoom,” or “don’t change the character,” and convert them into descriptive shot text that often improves adherence—without promising perfect control.

What “descriptive, not instructive” means (and why commands get ignored)

Poe’s documentation frames video prompting differently from text prompting: prompts should be descriptive rather than instructive, and even the “style prompt” should describe the desired video, not tell the bot what to do. (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)

A useful mental model:

Instructive prompts: written like directions to a crew (“don’t do X,” “add Y,” “make it cinematic,” “keep the character identical”).
Descriptive prompts: written like what the viewer will actually see and feel (“a handheld close-up with shallow depth of field,” “the same woman in a red raincoat,” “a slow push-in as tension builds”).

Poe also suggests a simple prompt structure for video:

Style
Camera angle
Description of character or scene + a verb
Background / environment / setting
Any additional information (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)

And importantly: keep prompts concise—Poe notes that shorter prompts tend to work better than longer ones for video generation bots. (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)

The 5 command-types creators should stop writing (and what to write instead)

These are the most common “instruction” buckets that cause drift or partial compliance. The fix isn’t magic words—it’s converting “orders” into “observable attributes.”

1) Vibe commands (“make it cinematic”, “make it epic”)

Rewrite into: lens/shot choices, lighting, texture, pace, and mood.

2) Camera commands (“do a dolly zoom”, “add a drone shot”)

Rewrite into: what the camera does as seen on screen.

Poe explicitly notes that adding camera motions can produce more cinematic results. (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)

3) Negative commands (“don’t change the character”, “no text”, “no watermark”)

Rewrite into: positive constraints and identity anchors (wardrobe, physical traits, props, environment consistency).

4) Editing/pacing commands (“cut faster”, “add a whip pan transition”)

Rewrite into: a single continuous shot, or clearly described beats over time.

QuestStudio summarizes common official guidance across tools: good prompts are clear about shot, motion, camera, and mood. (https://queststudio.io/blog/text-to-video-prompts)

5) “Do what I mean” meta-instructions (“keep it consistent”, “follow the reference exactly”)

Rewrite into: the specific details you need preserved (colors, clothing, location, time of day, composition).

Side-by-side: Poe vs Adobe vs FlexClip—then a Veo3Gen decision rule

Different docs emphasize different levers:

Poe: keep it descriptive, and shorter often works better for video prompts. (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)
Adobe: their guidance emphasizes being very specific about what you want (more detail tends to mean more words). (https://helpx.adobe.com/firefly/web/work-with-audio-and-video/work-with-video/writing-effective-text-prompts-for-video-generation.html)
FlexClip: offers a clear structure: Subject + Action + Scene + (Camera Movement + Lighting + Style). (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos)

A simple decision rule for Veo3Gen:

Start with Poe-short: one shot, one action, one setting, one camera idea.
If outputs drift, add Adobe-specificity only where it fails (identity anchors, environment, lighting).
Keep your prompt readable by following FlexClip’s order (Subject → Action → Scene → Camera/Light/Style). (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos)

Rewrite Table: 25 common “instructions” → descriptive shot text (copy/paste)

Use these as patterns. Left is what creators type; right is what the model can “paint.”

Command-style instruction	Descriptive rewrite (shot text)
“Make it cinematic.”	“Moody cinematic lighting, shallow depth of field, soft film grain, slow deliberate pacing.”
“Make it look like a blockbuster.”	“High-contrast dramatic lighting, wide establishing shot, crisp detail, subtle handheld energy.”
“Add a dolly zoom.”	“The camera pushes in while the background appears to stretch away, creating a tense vertigo effect.”
“Do a slow dolly-in.”	“Slow push-in toward the subject from a medium shot to a close-up.”
“Add camera shake.”	“Handheld shot with gentle micro-shake, as if filmed on a shoulder rig.”
“Do a drone shot.”	“Aerial overhead view gliding forward above the scene, revealing the location.”
“Pan left.”	“The camera pans left, revealing a new detail entering frame from the right.”
“Tilt up.”	“The camera tilts up from the subject’s hands to their face, then to the skyline.”
“Rotate the camera.”	“The horizon slowly rolls clockwise as the subject remains centered.”
“Zoom in.”	“A gradual zoom-in from wide to medium, keeping the subject centered.”
“Make it fast-paced.”	“Quick movement: the subject strides briskly; the camera follows tightly with energetic motion.”
“Slow motion.”	“The action unfolds in slow motion; dust particles and fabric movement are emphasized.”
“Make it scary.”	“Low-key lighting, deep shadows, narrow beam of light, tense stillness before a sudden movement.”
“Make it happy.”	“Bright soft daylight, warm color palette, relaxed body language, gentle upbeat mood.”
“Make it more dramatic lighting.”	“Hard key light from the side, strong shadows on the opposite cheek, high contrast.”
“Add neon lighting.”	“Neon signs cast magenta and cyan rim light across wet pavement and the subject’s jacket.”
“Golden hour.”	“Warm sunset light with long soft shadows; amber highlights on hair and skin.”
“Don’t change the character.”	“Same person throughout: woman with short black bob haircut, green hoodie, small scar on left eyebrow.”
“Keep the outfit the same.”	“The subject wears the same red raincoat and black boots in every moment of the clip.”
“No text on screen.”	“Clean frame with no on-screen typography, logos, or captions.”
“No watermark.”	“Unbranded, clean image with no visible overlays or marks.”
“Don’t change the background.”	“The scene stays in the same location: a small kitchen with blue tiles and a window on the left.”
“Make it photoreal.”	“Photorealistic look, natural skin texture, realistic reflections and shadows.”
“Make it anime style.”	“Anime-inspired illustration style, clean linework, cel shading, expressive eyes.”
“Add b-roll shots.”	“A sequence of detail shots: hands adjusting a dial, close-up of texture, then a wider shot showing the object in context.”

Tip: If your tool supports explicit camera motion parameters, Poe lists examples like --zoom (in/out), --rotate (cw/ccw), --tilt (up/down), and --pan (left/right). (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)

Camera motion the descriptive way (without “do X” phrasing)

You can imply motion by describing:

Start frame → end frame: “Starts on the empty hallway; ends on the character’s face in close-up.”
Tracking relationship: “Camera follows two steps behind the subject at walking speed.”
Reveals: “The camera slides right, revealing the product on a pedestal.”
Subject-driven motion: “The subject turns; the camera arcs to keep them in profile.”

This aligns with common guidance summarized by QuestStudio: strong prompts usually include shot type, camera movement, lighting/style, and motion over time (plus constraints that matter). (https://queststudio.io/blog/text-to-video-prompts)

Length control: when shorter prompts win vs when specificity wins

As of 2026-05-08, a practical way to reconcile “short prompts work” with “be very specific”:

Start short and descriptive (Poe): one shot, one action, one setting; fewer “director notes.” (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)
Add specificity only for failure points (Adobe): if the model changes identity, lighting, or setting, add anchors for those. (https://helpx.adobe.com/firefly/web/work-with-audio-and-video/work-with-video/writing-effective-text-prompts-for-video-generation.html)
Keep structure consistent (FlexClip): Subject + Action + Scene + (Camera Movement + Lighting + Style), with action as the “core” that drives the storyline. (https://help.flexclip.com/en/articles/10326783-how-to-write-effective-text-prompts-to-generate-ai-videos)

A/B test method in Veo3Gen: prove the rewrite helped in 10 minutes

Don’t guess—test.

Quick A/B checklist (creator-friendly)

Hold subject and scene constant (same identity anchors, same location).
Write A = command-style and B = descriptive shot text.
Generate 3 variations per prompt (A1–A3, B1–B3).
Score each result 1–5 on:
- Motion clarity (does movement read cleanly?)
- Camera compliance (does framing/motion match?)
- Drift (identity/wardrobe/setting changes?)
Keep the better prompt, then add one detail at a time if needed.

This approach matches the spirit of “clarify shot/motion/camera/mood” guidance across tools. (https://queststudio.io/blog/text-to-video-prompts)

Mini prompt kits: 6 ready-to-use descriptive prompts

Each kit follows Poe’s suggested structure (style → camera angle → character/scene + verb → background/setting → extra info). (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)

1) Product ad (single hero shot)

Prompt: “Clean commercial style, medium close-up at table height. A matte black water bottle rotates slowly as condensation beads catch the light. Minimal studio backdrop, soft gradient background. Softbox key light with gentle rim light, crisp reflections.”

2) App demo b-roll

Prompt: “Modern tech b-roll style, over-the-shoulder close-up. A person taps through a mobile budgeting app and scrolls smoothly. Cozy home office with a plant and a desk lamp in the background. Warm practical lighting, shallow depth of field.”

3) Talking head (creator intro)

Prompt: “Natural vlog style, eye-level medium shot. A creator speaks directly to camera, calm confident tone, subtle hand gestures. Simple neutral background with a bookshelf. Soft window light from the left, clean audio-ready framing.”

4) Documentary moment

Prompt: “Documentary realism, wide establishing shot. A lone hiker walks along a foggy ridge line. Mountain landscape with low clouds drifting behind. Muted colors, gentle wind movement, slow steady camera follow.”

5) Anime reaction shot

Prompt: “Anime illustration style, tight close-up. A teen character’s eyes widen, then they smile slightly. School hallway background with soft pastel tones. Clean linework, cel shading, subtle camera push-in.”

6) Food macro

Prompt: “High-end food cinematography, macro close-up. A knife slices through a ripe strawberry; juice glistens as seeds and texture stay sharp. Dark slate surface background. Strong side light, glossy highlights, slow motion feel.”

FAQ

What does “descriptive, not instructive” actually mean?

It means writing what should appear in the video (shot, action, setting, style) instead of giving the model commands like a director would. (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)

Should I keep prompts short or add lots of detail?

Poe notes shorter prompts often work better for video generation, but you can add specificity when you see consistent failure (e.g., identity drift). (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)

What’s the simplest prompt structure I can follow?

Poe suggests: style, camera angle, character/scene + verb, background/setting, then additional info. (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)

How do I express camera motion reliably?

Describe the visible camera behavior (push-in, pan reveal, tilt from hands to face) and/or use supported motion parameters where available; Poe lists examples like --zoom, --rotate, --tilt, and --pan. (https://creator.poe.com/docs/prompt-bots/best-practices-for-video-generation-prompts)

CTA: Put the rewrite playbook into production

If you’re ready to turn these descriptive prompt patterns into something repeatable in your pipeline, explore the Veo3Gen endpoints and options in the API docs. When you want to scale tests (A/B batches, variations, prompt libraries), review plans on our Pricing page.

Poe’s “Descriptive, Not Instructive” Video Prompt Rule—A Creator Rewrite Playbook for Veo3Gen (as of 2026-05-08)

Try Veo 3 & Veo 3 API for Free