How to write prompts for Veo 3.1 (examples, tips and full guide) - Blog

Veo 3.1 doesn't need a secret format. It needs you to think like a director.

Google's GitHub have their official prompting guide for Veo 3.1 (normal and lite), this article puts everything together, with real examples.

Try Google Veo 3.1 here and Veo 3.1 Lite here 🎬

The formula behind every good Veo 3.1 prompt

Google's guide gives one formula:

[Cinematography] + [Subject] + [Action] + [Context] + [Style & Ambiance]

Cinematography: the camera work and the shot composition. Wide shot, close-up, crane shot, tracking shot.
Subject: the main character or focal point.
Action: what the subject is doing.
Context: the environment and the background.
Style & Ambiance: the overall look, the mood and the lighting.

Unlike MiniMax H3, there are no field names to type. It's plain prose, in one block, and the order above is simply the order that works best. This is the example Google uses:

Medium shot, a tired corporate worker, rubbing his temples in exhaustion, in front of a bulky 1980s computer in a cluttered office late at night. The scene is lit by the harsh fluorescent overhead lights and the green glow of the monochrome monitor. Retro aesthetic, shot as if on 1980s color film, slightly grainy.

Medium shot (cinematography), a tired corporate worker (subject), rubbing his temples (action), a cluttered office late at night (context), retro aesthetic, 1980s color film, slightly grainy (style and ambiance). Five parts, three sentences.

Example with prompt (made with the Lite version):

Low-angle tracking shot, a teenage skateboarder in a faded red hoodie, carving down a steep empty street and kicking off the curb into a clean ollie, in a hillside neighborhood of pastel houses at golden hour. Long shadows stretch across the asphalt and the low sun flares through the trees. Shot on 16mm film with soft grain and warm, slightly washed-out colors. SFX: skateboard wheels rumbling on rough asphalt, the sharp clack of the tail hitting the ground. Ambient noise: distant dogs barking and a light breeze.

Camera language is your strongest tool

Google calls the cinematography part "the most powerful tool for conveying tone and emotion". Veo understands three groups of terms:

Camera movement: dolly shot, tracking shot, crane shot, aerial view, slow pan, POV shot, zoom shot.
Composition: wide shot, close-up, extreme close-up, low angle, high angle, eye level, worm's eye, two-shot, over-the-shoulder shot.
Lens and focus: shallow depth of field, deep focus, soft focus, wide-angle lens, macro lens.

The trick is to describe the move from start to finish, not just name it. Compare "crane shot of a hiker" with Google's own example:

Crane shot starting low on a lone hiker and ascending high above, revealing they are standing on the edge of a colossal, mist-filled canyon at sunrise, epic fantasy style, awe-inspiring, soft morning light.

The camera has a starting point (low, on the hiker), a direction (ascending) and a payoff (the canyon reveal). That's what makes it read like a shot instead of a keyword.

Lens terms work the same way. "Close-up with very shallow depth of field" tells Veo what to keep sharp and what to blur, which does more for mood than any adjective.

Directing the sound

Veo 3.1 always generates audio with the video, so if you don't describe the sound, the model invents it. Google's guide splits audio into three cues:

Dialogue: put the exact words in quotation marks. A woman says, "We have to leave now."
Sound effects: start with SFX: and describe them clearly. SFX: thunder cracks in the distance.
Ambient noise: start with Ambient noise: and describe the background. Ambient noise: the quiet hum of a starship bridge.

For conversations, the Gemini API docs show a screenplay style that works very well: the character's name, a short stage direction in parentheses, then the line in quotes.

A wide shot of a misty Pacific Northwest forest. Two exhausted hikers, a man and a woman, push through ferns when the man stops abruptly, staring at a tree. Close-up: Fresh, deep claw marks are gouged into the tree's bark. Man: (Hand on his hunting knife) "That's no ordinary bear." Woman: (Voice tight with fear, scanning the woods) "Then what is it?" A rough bark, snapping twigs, footsteps on the damp earth. A lone bird chirps.

Notice that the sounds sit at the end, after the action. Visuals first, then dialogue, then the sound bed.

One important note on language: Google says English is fully supported and other languages "may work but results can vary". Write the prompt in English. If you need dialogue in Spanish or another language, keep only the spoken line in that language, inside the quotes, and test it.

Example with prompt:

Medium two-shot, inside a tiny ramen shop at night, steam rising from the counter. An old chef in a white headband slides a bowl toward a young delivery rider who is still wearing her helmet. Chef: (Gruff, without looking up) "Eat first. Deliveries can wait." Rider: (Laughing, pulling off her helmet) "You say that every night." The camera slowly pushes in on the bowl as she lifts her chopsticks. Warm tungsten light, rain-streaked window behind them, neon reflections on the wet glass. SFX: broth bubbling, chopsticks tapping the bowl. Ambient noise: rain on the window and a radio murmuring in the kitchen.

Negative prompts: say what you want instead

There is no separate negative prompt using Google's Veo model, their advice is to describe the result you want instead of listing what you don't want.

Instead of: no man-made structures
Write: a desolate landscape with no buildings or roads

The second version gives the model a picture to aim for. The first one only names the thing you're trying to avoid, which can pull it into the shot.

Adjectives: allowed, but make them concrete

If you read our MiniMax H3 prompt guide, you'll remember the rule "kill your adjectives". Veo 3.1 is different. Google's own guide says to "use adjectives and adverbs to paint a clear picture", and its examples use words like melancholic, awe-inspiring and epic fantasy style.

The catch: an adjective only helps when it points at something the camera can show. Compare these two prompts from Google's docs.

Less detail:

The camera dollies to show a close up of a desperate man in a green trench coat. He's making a call on a rotary-style wall phone with a green neon light. It looks like a movie scene.

More detail:

A close-up cinematic shot follows a desperate man in a weathered green trench coat as he dials a rotary phone mounted on a gritty brick wall, bathed in the eerie glow of a green neon sign. The camera dollies in, revealing the tension in his jaw and the desperation etched on his face as he struggles to make the call. The shallow depth of field focuses on his furrowed brow and the black rotary phone, blurring the background into a sea of neon colors and indistinct shadows, creating a sense of urgency and isolation.

"Desperate" appears in both. In the second one it's backed by things you can see: tension in his jaw, a furrowed brow, a background blurred into neon. That's the difference.

Timestamp prompting: several shots in one clip

This is the most powerful technique in Google's guide. You split the clip into timed segments and give each one its own shot, action and sound:

[00:00-00:02] Medium shot from behind a young female explorer with a leather satchel and messy brown hair in a ponytail, as she pushes aside a large jungle vine to reveal a hidden path.

[00:02-00:04] Reverse shot of the explorer's freckled face, her expression filled with awe as she gazes upon ancient, moss-covered ruins in the background. SFX: The rustle of dense leaves, distant exotic bird calls.

[00:04-00:06] Tracking shot following the explorer as she steps into the clearing and runs her hand over the intricate carvings on a crumbling stone wall. Emotion: Wonder and reverence.

[00:06-00:08] Wide, high-angle crane shot, revealing the lone explorer standing small in the center of the vast, forgotten temple complex, half-swallowed by the jungle. SFX: A swelling, gentle orchestral score begins to play.

A few rules that keep it working:

The segments have to add up to your duration. Veo 3.1 makes clips of 4, 6 or 8 seconds, so four segments of two seconds is the maximum that still breathes. More than that and every shot turns into a blink.
Repeat the key description of your character in the first segment, so every later shot has something to stay consistent with.
Every new segment should show something new: a reverse angle, a reveal, a change of place. If you only want to get closer, use a camera move inside one segment instead of a new one.

Example with prompt:

[00:00-00:02] Extreme close-up of a barista's hands pouring steamed milk into a white espresso cup, forming a slow rosetta, in a sunlit corner café with wooden counters. SFX: milk hissing, a soft pour.

[00:02-00:04] Over-the-shoulder shot from behind the barista as she slides the cup across the counter to a man in a grey suit who is checking his watch. Ambient noise: low café chatter and cups clinking.

[00:04-00:06] Close-up of the man's face as he takes the first sip, closes his eyes and finally relaxes his shoulders. Man: (Quietly, almost to himself) "Okay. Now the day can start."

[00:06-00:08] Wide shot from outside the café window, morning light on the glass, the man sitting alone at the counter as the barista wipes the espresso machine behind him. Warm, natural color grading.

The ways to start a video on Artificial Studio

In the Create Video tool, Veo 3.1 has three starting points, plus References you can add to any of them. Each one changes what your prompt should say.

Text to video: only words. The model builds everything from your prompt. Use the full five-part formula.

Start image: your image becomes the first frame. Google's docs are clear that Veo uses the input image as the initial frame, so pick the image that's closest to how you want the video to open. Then write the prompt about what happens next: the motion, the camera and the sound. Don't describe the picture again, it's already there. And if you really don't know what to write, Google's notebook says you can leave the prompt empty and let the model decide.

This is how Google writes it. The image shows tiny surfers in a stone bathroom sink. The video prompt only adds motion and camera:

A surreal, cinematic macro video. Tiny surfers ride perpetual, rolling waves inside a stone bathroom sink. A running vintage brass faucet generates the endless surf. The camera slowly pans across the whimsical, sunlit scene as the miniature figures expertly carve the turquoise water.

Start image + Last frame: you set the beginning and the end. Veo creates the transition between them. Your prompt should describe the journey, the camera move and the audio, not the two pictures. Google's example uses a front shot of a singer as the first frame and a view from behind her, facing the crowd, as the last one:

The camera performs a smooth 180-degree arc shot, starting with the front-facing view of the singer and circling around her to seamlessly end on the POV shot from behind her on stage. The singer sings "when you look me in the eyes, I can see a million stars."

The last frame always needs a start image. You can't use only an ending.

💡 Tip: you don't need to have the images already. Generate both frames first with GPT Image on Artificial Studio, using the same character description in both prompts, then bring them into Create Video as the Start Image and Last Frame. This is exactly the workflow Google recommends.

Click here to try Veo 3.1 on Artificial Studio 🎬

References: keep a character, an object or a style consistent

A start image is a frame. A reference is different: it tells Veo what something looks like, and the model puts it into a brand-new shot. Google calls this "ingredients to video", and it's how you keep the same character, product or look across several clips.

Above the Start Image, the References section gives you four options:

Style: a visual style you saved before (a color palette, an illustration look, a film grade).
Character: a person or character you saved before, so they look the same in every video.
Object: a product, prop or any item you saved before.
Image: attach your own reference images right there, without saving them first.

Then point at each reference from the prompt with @. That's what tells Veo which reference plays which role in the scene. Write the prompt like the formula says, and use the @ mention where the subject would go:

Medium shot, @Image1 sits behind a cluttered desk in a 1940s office at night, rain running down the window behind him. He looks up as @Image2 walks in and says in a weary voice, "Of all the offices in this town, you had to walk into mine." Venetian blind shadows across the room, black and white film noir style. Ambient noise: rain against the glass and a ceiling fan turning slowly.

A few things that make references work:

One role per reference. Google's own advice is to use images of a single person, character or product, so each reference has one clear job.
Don't re-describe the reference. The image already says how the character looks. Spend your words on the action, the camera, the place and the sound.
Reuse the same references in every clip. Two shots of the same scene with the same @Character keep the face, the clothes and the style consistent, which is what makes a multi-clip story feel like one film.
Plan for 8 seconds. In Google's specs, videos made with reference images are 8 seconds long.

💡 Tip: if a character or product will show up again, save it as a Character or Object instead of uploading the image every time. Next time it's one click, and it always looks the same.

Example with prompt:

First frame prompt (for the image model):

Wide shot of an empty, dusty pottery studio at dawn, a potter's wheel in the center with a lump of wet grey clay on it, soft blue light through a dirty window. Photorealistic.

Last frame prompt (for the image model):

Close-up of a finished glazed ceramic vase in deep cobalt blue, sitting on the same potter's wheel in the same studio, now lit by warm golden afternoon light through the window. Photorealistic.

Veo 3.1 prompt:

A time-lapse that starts on the lump of clay and pushes in slowly toward the potter's wheel as the day passes. Hands appear and disappear in quick blurs, the clay rises and takes shape, the light through the window shifts from cool blue to warm gold, and the camera ends on a close-up of the finished cobalt vase. SFX: the soft whir of the wheel, wet clay squelching. Ambient noise: birds outside that grow louder as the morning goes on.

Settings that change the result

Setting	Options	What to know
Duration	4, 6 or 8 seconds	1080p and 4K only work with 8 seconds
Aspect ratio	Auto, 16:9 or 9:16	Use 9:16 for Reels, TikTok and Shorts
Resolution	720p, 1080p, 4K (Veo 3.1)	Veo 3.1 Lite goes up to 1080p

Veo 3.1 or Veo 3.1 Lite?

Veo 3.1: 80 credits per second at 720p and 1080p, 120 at 4K. An 8-second clip at 1080p costs 640 credits. Use it for the final version.
Veo 3.1 Lite: 10 credits per second at 720p, 16 at 1080p. An 8-second clip at 720p costs 80 credits. Use it to test your prompt, then move the prompt that works to Veo 3.1.

💡 Tip: Since both models read prompts the same way, testing on Lite first is the cheapest way to iterate.

Before and after

Here's a prompt most people would write:

A cute snow leopard walking in a forest, 3D cartoon.

And here's the same idea, rebuilt with the formula (based on Google's own snow leopard example):

Create a short 3D animated scene in a joyful cartoon style. A cute creature with snow leopard-like fur, large expressive eyes, and a friendly, rounded form happily prances through a whimsical winter forest. The scene should feature rounded, snow-covered trees, gentle falling snowflakes, and warm sunlight filtering through the branches. The creature's bouncy movements and wide smile should convey pure delight. Aim for an upbeat, heartwarming tone with bright, cheerful colors and playful animation.

Longer, yes. But now Veo knows the style (3D cartoon), the character design (eyes, shape, fur), how it moves (prances, bouncy), the setting (rounded trees, falling snow) and the light (warm sunlight through the branches). Add a camera move and an SFX: line and you have a complete prompt.

Let the AI write the prompt for you

Google's guide suggests using Gemini to take a simple idea and enrich it with cinematic language. Just use Claude or ChatGPT to make it.

You can do it all in one place with the Artificial Studio MCP. It connects Claude, ChatGPT, Claude Code, Cursor and other MCP clients to Artificial Studio, so your assistant can write the Veo 3.1 prompt, show you the cost and generate the video without leaving the chat.

To connect it:

Claude: go to Customize → Connectors → Add custom connector and paste https://api.artificialstudio.ai/mcp
ChatGPT: turn on Developer mode in Settings → Apps → Advanced settings, create an app and paste the same URL
Claude Code: run claude mcp add --transport http artificial-studio https://api.artificialstudio.ai/mcp

There's no API key, and you get 20 free credits every day. Then just ask something like:

Write a Veo 3.1 prompt using the five-part formula (cinematography, subject, action, context, style and ambiance) for an 8-second 9:16 video of a florist closing her shop at night. Add one line of dialogue in quotes, an SFX line and an Ambient noise line. Test it on Veo 3.1 Lite first, and if it looks good, generate it on Veo 3.1 at 1080p. Use this article as a reference.

Your assistant writes the prompt, asks you to confirm the credits and returns the video in the chat.

Going past 8 seconds

Veo 3.1 generates up to 8 seconds per clip on Artificial Studio, so longer pieces get built shot by shot:

Write each shot as its own prompt, using the same saved Character, Object and Style references every time
Repeat the same character and style description too, so the prompt and the references agree
Use the last frame of one clip as the Start Image of the next one, so the action continues where it stopped
Keep the same style line in every generation: same film look, same color, same light direction
Cut on a sound: a footstep, a door, the end of a sentence. Never in the middle of a word
Use one music track for the whole edit instead of the music each clip generates, so the piece doesn't sound like several rooms glued together

The quick checklist

Cinematography, subject, action, context, style and ambiance, all present
Camera moves described with a start, a direction and an end
Dialogue in quotes, with who says it and how
SFX: for effects, Ambient noise: for the background
What you don't want written as what you do want
Adjectives backed by something visible
Timestamps that add up to 4, 6 or 8 seconds
Start image? Describe the motion, not the picture
Start and last frame? Describe the journey between them
References? Point at each one with @ and describe the action, not the look
Prompt in English, only the dialogue in another language if needed
Tested on Veo 3.1 Lite before the final render

What makes Veo 3.1 different

Veo 3.1 is Google's dedicated video model, and its strength is cinematic realism: lighting that behaves like real light, camera moves that feel like real equipment, and audio generated in sync with the picture, from a single footstep to a two-person conversation.

That's why its prompts read like a director's notes rather than a technical file. You aren't filling in fields, you're describing a shot: where the camera is, who is in front of it, what they do, where they are, and what it all sounds like.