Veo 3.1 from Google is one of the top models for video generation along with Kling AI, SeeDance, HappyHorse and Grok Imagine. It works simply: you write in text what you want to see, and she puts together a finished video. And immediately with sound, replicas of the characters, environmental noise and music.
The finished video will be as correct as you can clearly describe the scene to it. The model does not guess thoughts, she reads the text. Therefore, everything is decided by the prompt - the description you give her.
Making such descriptions is not difficult if you know the diagram. Below we will analyze it step by step, provide a template and ready-made examples.
And if you don't want to assemble it by hand, there is a shortcut. In Telegram bot Marty go to the Text section and send the neural network a prompt template (it will be below) and ask it to collect a description for your idea.
This is done directly in the chat. The AI will ask a couple of clarifying questions and issue a ready-made prompt. All you have to do is return to the “Video” section in the same bot and insert the prompt.
Where to start: Video section
The path is short: open the bot, go to the “Video” section. The prompt is responsible for the content of the scene, and the buttons in the bot are responsible for the format and quality.
Model. Choose between Veo 3.1 Pro and Veo 3.1 Fast. Pro provides maximum quality and detail, this is an option for a high-quality video. Fast works faster and spends fewer tokens, making it convenient to run drafts. The logic is simple: catch the scene on Fast, finish on Pro.
Quality. 720p or 1080p. 1080p is full HD, take it for everything that viewers will see. 720p is enough for tests.
Size. 9:16 or 16:9. 9:16 is a vertical under Shorts, Reels and TikTok. 16:9 is horizontal for YouTube, website or presentation. Choose a format in advance, depending on the platform where you will post the video.
Style. Anime, artistic, realistic or your own. A quick way to define the overall look of a video without describing it in words. If you need something non-standard, you can create your own style - this is done in a couple of clicks.
Bring the image to life. A button for those who already have a ready-made picture. You upload an image, and the model sets it in motion. There will be a separate block below about this function.
A detailed analysis of each button showing the interface is available in the video training for this section right in both - study the video to better navigate the bot and learn about all the features of the section. And I’ll teach you prompt engineering here for free.
What does a strong prompt consist of?
A good prompt is assembled from several semantic blocks. It is not necessary to describe every detail, but the clearer the description, the more accurate the result.
Who's in the frame? Not just man, and an elderly sailor with a thick gray beard in a blue pea coat. The more specific the image, the less the model thinks for you.
What's going on. One clear action: stands at the helm and looks at the waves. Many actions at the same time will simply confuse the model; it is easier for her when there is one clear scene.
Where. Environment and background: on the bridge of an old fishing boat in a storm. The location sets half the atmosphere.
In what style? Cinematic, cartoonish, documentary, selfie format. You can describe it in words in a prompt, or you can choose a ready-made style in the bot interface.
How filmed. Camera shot and movement: close-up, camera slowly zooms in on the face. The list of cinematic terms will be below - in the cheat sheet.
Light and mood. Golden hour and warm sunset light. Or cold neon of the night city. Light changes the feel of a scene more than it seems. And it’s better to set the light parameters right away.
What do you hear? Noises, sounds, music. Veo 3.1 voices the video itself, and if you don’t say what it should sound like, it will come up with it for you. There is a separate section about sound below.
What they say. If there is speech in the scene, write a line or prohibit the neuron from generating unnecessary dialogues (and it can come up with them itself).
Prompt template
In order not to keep the diagram in your head, collect the prompt using a template. Substitute your details instead of the hints in brackets and paste it into the bot in the “Text” section:
[who is in the frame, with a description] [what he is doing], [where, in what environment]. [visual style and atmosphere, lighting]. [shot and camera movement]. [sound: noises, music]. [if speech is needed: the hero speaks the required language, then a replica].
A few rules. Write in short phrases separated by a comma or period, this makes the prompt easier to read and edit. Do not indicate technical parameters: you have already set the duration, resolution and format using buttons. Avoid unnecessary symbols and em dashes; write numbers in words. Specify the language of speech explicitly: you need Russian, so write it in Russian, and then the replica.
Bad prompt - good prompt
I’ll show you one example so that it’s immediately clear why details make all the difference.
Lazy prompt: a man walks through the city.
The same plot according to the template: a young man in a long dark coat quickly walks along the evening street of the old city past glowing shop windows. Cinematic, soft light from lanterns, light rain glistening on the asphalt. The camera moves from the side and keeps the hero in a medium shot. Sound: footsteps on the wet pavement, the quiet hum of the city in the distance, without music.
In the second option, you control every element of the frame. The model does not guess, but fulfills your specifications.
Sound and speech
This is where newbies stumble most often. Veo 3.1 generates sound along with the picture, and if you don’t set it correctly, the model will add something of its own, sometimes inappropriately: the voices of a non-existent crowd can burst into a quiet scene. Record the sound as carefully as the picture.
Background noise. Say what should be heard around: the sound of the surf and wind, the roar of the cafe and the clink of dishes, the silence of an empty room.
Extra votes. If no one says so, write: without outside voices. This eliminates the random chatter and studio laughter in the background.
Music. Needed - indicate genre and mood: quiet sad piano melody, cheerful pop track. Not needed - write no music.
Replicas. Write the specific phrase after the colon: the old man says hoarsely: The sea teaches humility, one wave at a time. If the phrase is not important, describe the intent: old man shares a story about the sea, and the model will come up with a replica itself.
Replica length. The video is short, about eight seconds, and the phrase must fit within this time. The hero will jabber during a long monologue; one short sentence will suffice.
Trick against subtitles. Veo 3.1 likes to put text on the video when the hero speaks. To avoid this, add a note to the end of the prompt no subtitles.
Accent and intonation can also be stated in words: speaks in a soft calm voice. The model will take this into account.
Bring an image to life
If you already have a finished picture, you don’t need to describe the scene from scratch. The "Live Image" button will take your photo and make it move.
The logic of the prompt is different here: the scene is already in the picture, your task is to say what should move. You can say nothing at all, and the picture will still come to life.
Describe the movement. What exactly comes to life: hair flutters in the wind, a car drives forward, a person turns his head, waves move across the water.
Describe the camera. Light touch on the face, slow panorama, static frame without movement.
Add sound. Everything is the same as in a regular prompt: noise, music, a short remark if desired.
Don't overdo it. The fewer changes you ask, the more natural the result. Too many movements at once and the picture may float. One careful movement is better than ten sudden ones. Or don’t write anything at all - just click the “Live Image” button, upload your photo and click “Submit”.
Ready-made prompt examples
Several ready-made prompts for different tasks. Take them as a basis and substitute your own details.
Artistic frame. A storm-battered captain with a gray beard in a blue pea coat stands on the deck and points to the stormy waves. Close-up, camera slowly zooming in on the face. The golden light of the sunset through the storm clouds emphasizes wrinkles and harsh shadows. Sound: the howl of the wind and the sound of waves, without music. The captain speaks hoarsely in Russian: The sea teaches humility, one wave at a time. (no subtitles)
Product advertising. An elegant smartwatch rotates slowly on a glass stand in a bright, minimalist studio. The camera makes a smooth circular flight. Bright diffused light and a pure white background, soft highlights on the body. Sound: calm modern music, without extraneous noise. White, silver and accent blue predominate. (no subtitles)
Unboxing for the blog. A young blogger in a cozy room with posters on the walls sits in front of the camera and holds a box with a new gadget. Static medium shot, like when shooting on a webcam. Soft warm light from a table lamp. The blogger speaks animatedly in Russian: Hello everyone, today we will unpack this new product. Sound: light background music, quiet room noise. (no subtitles)
All three prompts are collected according to the same scheme: who is doing what, where, in what style, how it was shot, what kind of light and what is heard. Master this technique once, and then the prompt will be collected almost automatically.
Common mistakes and how to fix them
The most common mistakes of beginners. Go through the list before you start.
Too much action in one scene. The hero runs, jumps, talks and juggles, and the model gets confused. Leave one main action.
Forgotten sound. Didn't describe the audio, got random noises. Always add a block about sound, even if it is a video without music.
Overloaded replica. A long phrase does not fit into eight seconds. Keep it to one sentence.
A mess of styles. Cinematic, cartoonish and like an old movie all at once. Choose one direction.
Technical parameters in the text. There is no need to write the format, duration and resolution in words, everything is set using buttons.
If you don't like the result, change one parameter at a time. Don't rewrite the entire prompt, otherwise you won't understand what exactly worked. If you don't like the camera, just correct the line about the camera. The face floats, add details of appearance. The sound is out of place, just rewrite the block about the sound.
And don’t expect perfection the first time, there may be several runs. Run drafts on Fast so as not to burn a lot of tokens, and collect the final on Pro.
Cheat sheet on terms
A short reference book of words that the model understands well. Feel free to insert them into the prompt.
Plans:
General plan - the object is visible in its entirety along with its surroundings.
Medium shot - approximately waist-deep.
Close-up - a face or important detail is close.
First person - view through the eyes of the hero.
Angles:
At eye level - neutral look.
From above - The camera looks down on the stage.
From below - the camera looks up, the hero looks more impressive.
Camera movement:
Arrival and departure - the camera moves closer or further away.
Panorama - The camera turns in place, examining the scene.
Span - the camera flies over the stage as if from a drone.
Tracking - the camera moves next to the hero and keeps him in the frame.
Light:
Golden hour - warm soft light at sunset or dawn.
Neon - cold colored light of the night city.
Backlight - a source behind the hero, an expressive silhouette.
Soft diffused light - even lighting with almost no shadows.
What's the result?
The more clearly you describe the scene, the closer the video will be to what you originally intended. Prompt according to the scheme “who is doing what, where, how it was filmed, what kind of light, what can you hear” - this is already half the success, the rest is achieved through runs.
Now it's a matter of practice. Open the “Video” section in Marty, select the model and format, insert the prompt according to the template and run it.
The first video may not be perfect, and that's okay. Correct one parameter, run it again, and with each approach the result will be closer to the intended one.
Happy generation!