Veo Negative Prompts: How to Stop Unwanted Subtitles, Text, and Warped Faces

A Veo negative prompt works best when you describe what should be present instead of only saying what to remove — “no subtitles” alone is a weaker instruction than “no subtitles, no captions, no on-screen text, clean frame.” This guide covers the exact negative-prompt phrasing that suppresses Veo’s most common unwanted additions (subtitles, watermarks, warped faces) and what to do when a prompt still doesn’t produce what you asked for.

Why “no X” alone is a weak instruction

Diffusion and video models are trained to generate what’s described, not to subtract from a mental image. When you write only “no subtitles,” the model still has to interpret “subtitles” as a concept before it can (partially) avoid it — and the instruction competes weakly against everything else in the prompt describing what a talking-head video typically looks like (which, in a lot of training data, includes captions). The fix is to combine the negative with a positive description of the clean result you want: “no subtitles, no captions, no lower-third text — clean, unobstructed frame.”

The most common Veo problems and fixes

1. Burned-in subtitles / captions appear uninvited

Symptom: Dialogue or talking-head prompts often generate with auto-captions baked into the video, especially when native audio/dialogue is used.

Fix: Add an explicit, repeated negative line at the end of the prompt:

No subtitles, no captions, no closed captions, no lower-third text,
no on-screen text of any kind. Clean frame.

This is the highest-value negative prompt in this whole guide — it’s the single most-searched Veo problem. See it applied in the no-subtitles talking-head prompt in the Veo Prompt Library.

2. Watermarks or logos appear that you didn’t ask for

Fix:

No watermark, no logo, no brand marks, no UI overlay.

Pair this with a positive description of a “clean” surface if the shot includes blank space that might otherwise get filled with invented text.

3. Warped or morphing faces, especially in orbit or multi-person shots

Symptom: Faces distort mid-shot, especially during camera moves that circle a subject or in scenes with more than one person.

Fix: This isn’t purely a negative-prompt problem — it’s often a camera or complexity problem (see the camera movement guide). Two levers:

Shorten the camera move (a 90° arc instead of a full 360° orbit).
Add an explicit consistency instruction: “no face morphing, keep facial features consistent throughout.”

If you’re animating from a reference photo, see image-to-video prompts — letting the source image carry identity (rather than re-describing the face in text) reduces drift significantly.

4. Veo ignores a specific detail you asked for

Symptom: You specify a precise color, object, or action, and Veo substitutes something close but wrong.

Fix: This is usually a prompt-priority problem, not a negative-prompt problem. Move the detail earlier in the prompt (front-loaded tokens get more weight), and make it concrete rather than abstract — “matte-black ceramic mug” beats “a nice mug,” as covered in the prompt formula guide.

5. Generic “spooky” or “epic” mood words produce a flat result

Fix: Not a negative prompt — the actual fix is to stop using mood adjectives and describe real, physical, named sounds and visuals instead. “Spooky sounds” → “a distant floorboard creak, faint wind through a cracked window.” See Part 6 of the prompt formula guide.

6. Background music appears when you didn’t want it

Fix:

No background music, no score. Ambient sound only: [name the specific
ambient sound you do want].

Naming the ambient sound you do want alongside the negative gives Veo something concrete to render instead of silence-by-omission, which produces a more natural result than an empty negative.

7. Aspect ratio or duration comes out wrong

Fix: State the aspect ratio and duration explicitly and near the end of the prompt: “8 seconds, vertical 9:16.” This isn’t strictly a negative prompt, but it’s the same category of problem — an unstated technical spec defaults to whatever’s most common in training data (usually 16:9), so always state it if you need something else.

Negative prompt checklist (copy this into any Veo prompt)

No subtitles, no captions, no on-screen text, no watermark, no logo,
no lower-third, clean frame.

Add this as a standing line at the end of prompts for talking-head, dialogue, and UGC-style content — the categories where unwanted text and watermarks show up most often.

Before / after: a full negative-prompt pass

Before:

A founder in a studio talks to the camera about the product. No subtitles.

After:

Close-up talking-head video of a founder in a small studio looking into
the camera and saying, "The product is simple because the workflow is
not." Calm, confident delivery, natural lip sync, soft key light, subtle
room tone. Do not show subtitles, captions, lower thirds, logos, or any
on-screen text. 8 seconds.

This exact after-prompt is verified in the Veo Prompt Library.

FAQ

Do negative prompts work the same on Veo 3.1 and Veo 3.1 Lite? Broadly yes, though Lite is somewhat less consistent at fully suppressing subtitles on longer or busier dialogue scenes — expect to regenerate more often on Lite if captions keep appearing.

Why does Veo sometimes ignore my negative prompt entirely? Usually because the negative is a single short line competing against a long, detailed positive description. Try shortening the positive prompt or moving the negative earlier if it keeps getting dropped.

Is there a real “negative prompt” field in Veo, like Midjourney’s --no? Not as a separate parameter in the consumer Veo tools (VideoFX, Gemini, AI Test Kitchen) as of July 2026 — negatives are written inline in the same text prompt, phrased as “no X, no Y” or “do not show X.” Some API-level access may expose more controls; verify against current Vertex AI docs before assuming a dedicated parameter exists.

My prompt “isn’t working” — where do I start troubleshooting? In order: (1) check you’re describing the subject concretely, not vaguely; (2) check you’re using one camera move, not stacked; (3) check your negative prompt is specific (“no subtitles, no captions, no on-screen text”) not generic (“no text”); (4) check technical specs (aspect ratio, duration) are stated explicitly.

Where to test fixes quickly

Regenerating to test a negative-prompt fix costs credits — comparing behavior across models can help you learn faster which fixes are Veo-specific vs. universal. Pollo AI runs Veo alongside other models from one dashboard, useful for this kind of side-by-side troubleshooting. This is an affiliate link — we may earn a commission at no extra cost to you.

The Veo Prompt Builder automatically appends the standing “no subtitles, no watermark” negative line to dialogue and UGC presets, so you don’t have to remember it by hand.