Veo 3.1: how to write prompts the model actually understands

What is new in Veo 3.1

Veo 3.1 is five concrete upgrades on top of Veo 3. First: improved prompt adherence — the model follows descriptions more precisely, invents less. Second: native vertical 9:16 support for TikTok, Reels, and Shorts — no more cropping in post. Third: image-to-video — the model animates a starting frame, with the prompt describing motion, not the frame itself.

Fourth: camera presets — built-in movement presets (platform-specific) that supplement the text description. Fifth: longer clips compared to the 8-second Veo 3 ceiling. Base Veo 3.1 resolution is up to 1080p; the Fast and Fast Relax variants run at 720p with higher speed and lower cost. Prompt limit grows to about 2000 characters.

• Up to 1080p (Veo 3.1), 720p (Fast / Fast Relax)
• Native 9:16 — TikTok, Reels, Shorts without cropping
• Image-to-Video: animate a starting frame
• Camera presets + extended duration
• Prompt limit ~2000 characters, audio inherited from Veo 3

Prompt structure

Optimal order: [Subject + Appearance] + [Context/Scene] + [Action] + [Camera Movement/Composition] + [Style/Mood] + [Lighting] + [Dialogue: text] + [Audio/Ambiance] + [(no subtitles!)].

Veo 3.1 fully inherits Veo 3 audio logic: the sound block is mandatory, dialogue uses a colon, append `(no subtitles!)` at the end. Thanks to stricter prompt adherence the tag works more reliably than in Veo 3.

For complex scenes use a structured prompt with explicit blocks:

Scene: A busy cafe in Paris, morning light streaming through large windows. Character: A young woman with auburn hair, wearing a cream sweater. Action: She lifts a cup of coffee, takes a sip, looks out the window. Camera: Slow dolly-in from medium shot to close-up. Audio: Ambient cafe sounds, clinking cups, soft jazz piano. Mood: Warm, nostalgic, golden hour tones. (no subtitles!)

The model reads this layout better than a single long paragraph.

Vertical video and image-to-video

For vertical 9:16 the format is chosen in the platform (Google AI Studio, Vertex AI), not in the prompt. Adapt the prompt to a portrait composition: more close-ups, portrait orientation for the subject, minimal wide landscape shots (they get lost in 9:16). Selfie style fits the vertical format particularly well.

For image-to-video the model uses the uploaded image as the first frame, and the prompt describes motion and action, NOT the original frame. Weak: «A woman in a cafe drinking coffee» (that is already shown in the photo). Strong: «The woman slowly lifts the cup to her lips and takes a sip. Camera: slow dolly-in to extreme close-up on her eyes. Background Sound: faint cafe chatter, distant espresso machine.». Description of the initial state is just noise — focus on motion only.

Dialogue, audio, subtitles

Veo 3.1 fully inherits Veo 3 audio capabilities. Dialogue via colon, not in quotes: `says: text` works better than `says "text"` — quotes trigger embedded subtitles. Append `(no subtitles!)` to the end of the prompt.

Dialogue length must fit the clip duration: roughly 12-25 words per 8-second take. Too long — the model speaks unnaturally fast. Too short — it fills pauses with AI mumbling. With multiple characters state clearly who is speaking: «The woman in red says: ... The man with beard replies: ...».

Write background sounds out explicitly — even a plain «ambient room tone» removes the risk of «studio audience laughter». For music specify genre and mood: «a melancholic orchestral score swells», «upbeat electronic music with a driving beat», «no background music — just ambient room tone». Veo 3.1 follows these instructions more precisely than Veo 3.

Common mistakes

1. 1. Describing the starting frame in Image-to-Video

   In image-to-video mode the image ALREADY locks the first frame. Writing «A woman sitting in a cafe drinking coffee» in the prompt is just empty repetition of what the photo already shows. Describe MOTION only: «She slowly lifts the cup, takes a sip, looks out the window. Camera: slow dolly-in». Focus on dynamics, not statics.

2. 2. Format specified in the prompt text

   «Vertical video», «9:16», «1080p» in the prompt text are ignored — these are generation parameters set on the platform or via the API. In the prompt they become noise. If you need vertical, pick it in Google AI Studio / Vertex AI and adapt the composition: «portrait close-up», «subject centered», close shots.

3. 3. Horizontal composition with 9:16 selected

   If the vertical format is selected but the prompt still says «wide establishing shot of a city skyline» — the subject will be cropped and the frame loses meaning. For 9:16 adapt the composition: more close-ups, portrait orientation for people, minimal wide landscapes. Selfie style fits the vertical format particularly well.

4. 4. Dialogue in quotes without «no subtitles»

   Veo 3.1 inherits Veo 3 subtitle behavior: quotes around dialogue trigger embedded captions at the bottom of the frame, often with typos. Use the `says: text` format with a colon and append `(no subtitles!)`. In Veo 3.1 the tag works more reliably than in Veo 3 thanks to improved prompt adherence.

5. 5. Re-rolling at an identical prompt

   Veo 3.1, like Veo 3, is very consistent — an identical prompt yields a similar result. For variation CHANGE the prompt: swap the lens, change the lighting, alter the color palette, add a character detail. Re-rolling without changes wastes tokens; real variation comes only from edits.

Before / after examples

a woman drinking coffee in a cafe

Scene: A busy cafe in Paris on a Saturday morning, golden light streaming through large arched windows.
Character: A young woman with auburn hair tied in a low ponytail, wearing a cream cable-knit sweater, sits at a small marble table by the window.
Action: She lifts a small espresso cup, takes a slow sip, then sets it down and looks out the window with a pensive expression.
Camera: Slow dolly-in from medium shot to close-up on her face.
Lighting: Warm golden hour light through the windows, soft fill from a nearby lamp.
Mood: Warm, nostalgic, contemplative.
Audio: Ambient cafe sounds — clinking cups, soft conversation in French, a jazz piano playing quietly in the background.
(no subtitles!)

vertical video of a person in the city

Vertical 9:16 composition optimized for mobile. A young man with messy dark hair and a black hoodie, leaning against a graffiti-covered wall in a neon-lit Tokyo alley. He looks down at his phone, smiles, then glances up at the camera. Camera: portrait close-up, slight handheld micro-shake, slow push-in. Lighting: cyan neon key from screen-left, warm spill from a noodle shop sign on screen-right. Style: slightly grainy, film-like, cinematic vlog aesthetic. Mood: cool, urban, intimate. Background Sound: distant traffic hum, faint J-pop playing from a nearby shop, light rain on metal awnings. (no subtitles!)

animate this product photo of headphones

[Image-to-Video: starting frame is a product shot of matte-black wireless headphones on a white marble pedestal]

The headphones begin a slow, smooth 360-degree rotation on the pedestal. Camera: slow continuous orbit around the headphones at eye level, shallow depth of field maintained throughout. Lighting: existing softbox key and rim light from the starting frame, with subtle highlight movement as the headphones rotate. Style: clean commercial photography. Mood: premium, refined. Audio: subtle electronic ambient tone, soft mechanical hum, a gentle chime at the start of rotation. (no subtitles!)