Why “positive + negative” can beat longer prompts (and what to avoid)

Most creators don’t lose time because they lack words—they lose time because their prompt contains conflicting instructions.

Luma’s guidance for Dream Machine is an instructive cautionary tale here: it recommends a positive-only approach for optimal results, emphasizing that you should clearly describe what you want the model to generate. (https://lumaai-help.freshdesk.com/support/solutions/articles/151000219614-understanding-prompting-for-dream-machine-positive-vs-negative)

It also explains what negative prompting does—it tells the AI what to exclude—and warns it can be counterproductive: telling the model “don’t include X” can lead it to think about X, introduce it, then attempt to remove it, which can increase the chance of unwanted elements appearing. (https://lumaai-help.freshdesk.com/support/solutions/articles/151000219614-understanding-prompting-for-dream-machine-positive-vs-negative)

So why use a “POSITIVE + NEGATIVE” structure at all in Veo3Gen?

Because in real creator workflows (shorts, ads, UGC, b-roll), you often need a tiny guardrail list—not a giant “ban list.” The trick is to:

Make the POSITIVE block do 90% of the work.
Use short, specific NEGATIVE lines only for recurring deal-breakers (accidental logos, gibberish text, extra hands, etc.).
Avoid long negatives that over-constrain and make output bland or unstable.

If you remember nothing else: write what you want first; only then add a minimal “don’t” list.

The Veo3Gen “Do/Don’t” prompt template (copy/paste)

Use this two-block structure every time:

POSITIVE (what must appear)

Write in natural language—think of it as a short creative brief. Luma’s best practices describe prompting as using natural language and treating it like a conversation, which is a good default mindset for video prompting too. (https://lumalabs.ai/learning-hub/best-practices)

A simple ordering that stays readable is:

Shot/camera
Main subject
Action
Environment
Lighting/mood
Style constraints (only if needed)

One article breaks down a prompt element order as: [Camera type/shot], [Main subject], [Subject action], [Camera movement], [Lighting], [Mood]. (https://filmart.ai/luma-dream-machine/)

NEGATIVE (what must not appear)

Keep this short. Don’t list everything you dislike in the universe—only the top problems you keep seeing.

Copy/paste template:

POSITIVE:
[Shot + subject + action]. [Setting]. [Lighting/mood]. [Style constraints].

NEGATIVE:
No extra people. No extra limbs/fingers. No text or logos. No flicker.

Checklist: 8 common problems + exact negative lines to add

Here’s a short checklist you can keep next to your prompt box. Add only the lines you need.

1) Hands/fingers drift or mutate

NEGATIVE add-ons:

No extra fingers. No extra hands. No fused fingers.

POSITIVE reinforcement (optional):

“Hands remain natural and consistent throughout the shot.”

2) Faces morph between frames

NEGATIVE add-ons:

No face warping. No facial distortion.

POSITIVE reinforcement (optional):

“Same person, consistent facial features across the full clip.”

3) Unwanted text overlays appear (gibberish subtitles, random signage)

NEGATIVE add-ons:

No subtitles. No on-screen text. No watermarks.

If you do want text, be explicit. Luma’s best practices note you can request text by specifying wording like a poster with text that reads a certain phrase. (https://lumalabs.ai/learning-hub/best-practices)

4) Accidental logos/brands show up

NEGATIVE add-ons:

No logos. No brand marks. No trademarked designs.

5) Flicker / strobing / exposure pumping

NEGATIVE add-ons:

No flicker. No strobing. No exposure pulsing.

POSITIVE reinforcement (optional):

“Stable lighting and exposure.”

6) Style drift (turns into anime, clay, hyper-real, etc.)

NEGATIVE add-ons:

No animation style shift. No cartoon look. No CGI/plastic skin.

POSITIVE reinforcement (recommended):

Name your intended look once: “Natural smartphone video” or “cinematic live-action look.”

7) Extra objects appear (mystery props, extra products)

NEGATIVE add-ons:

No extra objects. No duplicate products. No clutter.

POSITIVE reinforcement (recommended):

“Only one [product] on the table.”

8) Weird camera behavior (unwanted zooms, spins, Dutch angles)

NEGATIVE add-ons:

No sudden zoom. No camera spin. No Dutch angle.

POSITIVE reinforcement (recommended):

“Locked-off shot” or “slow push-in” (pick one).

5 Veo3Gen prompt examples (POSITIVE + NEGATIVE)

These are designed for typical creator outputs: product shots, talking head, UGC ads, b-roll, and motion-graphic-style content.

Example 1: Clean product shot (ecommerce / Amazon-style)

POSITIVE:
Medium close-up product shot of a matte black water bottle standing on a white tabletop. Soft window light from the left, gentle shadows, minimal background. Slow, subtle push-in.

NEGATIVE:
No logos. No text. No extra objects. No reflections showing people.

Example 2: Talking head (creator intro for shorts)

POSITIVE:
Front-facing talking head, creator speaking to camera in a cozy home office. Natural skin texture, realistic smartphone video look, steady framing, friendly expression.

NEGATIVE:
No face warping. No extra teeth. No subtitles. No watermarks.

Example 3: UGC-style ad (hands demoing product)

POSITIVE:
Overhead shot of hands demonstrating how to use a small kitchen gadget on a wooden counter. Bright natural daylight, crisp focus on the product, simple background.

NEGATIVE:
No extra fingers. No extra hands. No duplicate product. No text or logos.

Example 4: Cinematic b-roll (brand film vibe)

POSITIVE:
Cinematic b-roll of a runner tying shoelaces at sunrise in a quiet city street. Warm golden light, shallow depth of field, slow handheld micro-movement, calm mood.

NEGATIVE:
No style shift to animation. No flicker. No random people crossing frame.

Example 5: Motion-graphic-style (clean shapes, minimal scene)

POSITIVE:
Minimal motion-graphic look: simple geometric shapes forming a clean title card background. Smooth easing, flat colors, high contrast, centered composition.

NEGATIVE:
No photoreal textures. No clutter. No random text.

How to keep negatives from fighting your positives (priority, specificity, ordering)

Negative prompts can help, but Luma’s help center warns they can also backfire by increasing the likelihood of the very elements you’re trying to avoid. (https://lumaai-help.freshdesk.com/support/solutions/articles/151000219614-understanding-prompting-for-dream-machine-positive-vs-negative)

To reduce that risk in Veo3Gen, use these rules:

1) Prioritize with your POSITIVE block

If the model keeps adding people, don’t only write “No people.” Instead, describe the actual target.

Luma’s article gives examples of translating negatives into positives, like replacing “a room, do not add people” with a clearer target like “an empty, minimalist room with natural light.” (https://lumaai-help.freshdesk.com/support/solutions/articles/151000219614-understanding-prompting-for-dream-machine-positive-vs-negative)

Same with “no cars”: specify “a pristine, untouched natural landscape…” rather than fighting the model with prohibitions. (https://lumaai-help.freshdesk.com/support/solutions/articles/151000219614-understanding-prompting-for-dream-machine-positive-vs-negative)

2) Make negatives short and concrete

Good: No subtitles. No watermark.

Risky: No bad quality, no ugly, no weird, no artifacts, no anything I don’t like...

Long “ban lists” tend to collide with your intent and can make results feel generic.

3) Don’t negate core features

If your positive says “smiling person speaking,” don’t add No teeth or No mouth movement.

If you need to fix something, rewrite the positive instead: “subtle smile, natural teeth, calm mouth movement.”

Mini case study: messy prompt → clean 10–12s ad shot in 3 iterations

Exact clip duration and behavior varies by model/version and settings; treat this as a workflow pattern (as of 2026-03-04).

Iteration 1 (too vague)

Prompt: “Make a cool ad video of my skincare bottle on a table, cinematic, no text, no logos, no weird stuff.”

What happens: The model may still invent label text or add extra props because “cool ad video” is underspecified.

Iteration 2 (tight positive, minimal negative)

POSITIVE:
Cinematic product shot of a plain white skincare bottle centered on a clean stone countertop. Soft spa-like daylight, neutral color grade, slow push-in, minimal scene.

NEGATIVE:
No text. No logos. No extra objects.

Why it improves: The positive specifies composition, surface, lighting, and camera move; negatives only block deal-breakers.

Iteration 3 (target the last 10%)

If you still see flicker or prop duplication:

NEGATIVE:
No text. No logos. No extra objects. No duplicate bottle. No flicker.

That’s it—resist the urge to add 40 more constraints.

Troubleshooting when negatives backfire

If you add negatives and the result gets worse (or bland), try this sequence:

Symptom: The model keeps showing the banned element anyway

Rewrite the POSITIVE to describe the desired alternative more vividly.
Replace “don’t add people” with “empty” / “unoccupied” environments (mirroring Luma’s positive rewrite examples). (https://lumaai-help.freshdesk.com/support/solutions/articles/151000219614-understanding-prompting-for-dream-machine-positive-vs-negative)

Symptom: Output becomes generic or lifeless

Remove broad negatives like No stylized or No dramatic lighting.
Add one positive style anchor: “natural handheld smartphone look” or “cinematic live-action.”

Symptom: Negatives conflict with the main subject

If your product has a logo you actually need, don’t write No logos.
Instead: “Logo is clear and correctly spelled” (and remove the negative).

Symptom: Inconsistent results across tries

Keep the prompt stable and change one variable at a time.
Consider a “board-like” iteration workflow; Dream Machine, for example, retains context within a board and “remembers” earlier generations, which encourages iterative refinement. (https://lumalabs.ai/learning-hub/best-practices)

Quick FAQ

Should I always use negative prompts in Veo3Gen?

No. Use them only when you have repeat offenders (text overlays, extra limbs, logos). Luma’s guidance for Dream Machine recommends positive-only for optimal results, and also warns negatives can increase unwanted elements. (https://lumaai-help.freshdesk.com/support/solutions/articles/151000219614-understanding-prompting-for-dream-machine-positive-vs-negative)

When should I rewrite the POSITIVE instead of adding more negatives?

When the issue is “the model doesn’t understand what I want.” If you’re trying to ban people/cars/clutter, describe “empty,” “pristine,” “minimalist,” etc., similar to Luma’s examples. (https://lumaai-help.freshdesk.com/support/solutions/articles/151000219614-understanding-prompting-for-dream-machine-positive-vs-negative)

Explore the docs and try your first integration: /api
See plans and usage options: /pricing

Keep your POSITIVE block crisp, your NEGATIVE block short, and iterate like a creator—not like you’re writing a legal contract.

Try Veo3Gen (Affordable Veo 3.1 Access)

If you want to turn these tips into real clips today, try Veo3Gen:

Start generating via the API: /api
See plans and pricing: /pricing

Positive vs Negative Prompts for AI Video (Veo3Gen Edition): A Simple “Do/Don’t” Checklist That Fixes Drift, Hands, and Unwanted Style

Try Veo 3 & Veo 3 API for Free