Lesson 2 · 17 min
A video prompt is the shot-list line written for the machine. Who it is, what they do, where, how the camera sees it, what light, and what you hear. Now that sound comes out together with the image, it goes in the prompt too, in a separate sentence.
Start the lesson
Goal: write the prompt for a shot in seven parts, with the sound in a separate sentence and what to avoid described without "no".
Not every prompt uses all of them, but each one answers a question. Tap each part.
Practical order: what you see first (subject, action, setting), then how you see it (camera, light), then what you hear.
The example is shot 4 of "The first batch": the oven door opens and the steam rises.
Start with what a viewer would understand in one second: who, doing what, where. Swap generic adjectives for detail that only exists in this scene.
A living verb carries the story: "the iron door opens slowly and the steam rises against the light" says more than "an oven in a bakery".
"bakery oven" → "the iron door of a deck oven opens slowly and the steam rises against the light"
Size, angle and movement are on the shot's line in your shot list. The light is on the look card. Copy, don't rewrite: that's what keeps the scene consistent.
If you don't have these files yet, do lessons 2 to 4 and 7 of course 0 first.
4 | close-up | low angle | dolly in | 4 s → "Close-up, low angle, slow dolly in."
Google's official video prompt guide (Veo and Gemini Omni Flash, updated on 10/15/2026) recommends saying clearly that you want audio and describing it in separate sentences: effects, ambient sound and dialogue.
The guide, in English, writes who speaks, "says:" and the text. In Portuguese, this course uses "who says: text". A short sentence fits in the shot; a paragraph doesn't.
Three knocks on the iron door. The baker says: Coming! In the background, the radio, low.
In the negative prompt field, list what you don't want as things: "on-screen text, watermark, people in the background". The same guide advises against "no" instructions, like "don't show text".
There's no consensus on the ideal prompt length: sources range from 10 to 150 words. Start complete and cut what the generator ignores.
The prompt sheet that the page checks, and the request for the AI assistant to put everything into one flowing text.
Subject: <who or what, with detail from the character sheet> Action: <what happens, with a lively verb> Setting: <where and when, with one detail unique to that place> Camera: <size, angle and one movement, from the shot list> Light: <your look card> Sound: <ambient sound, effect and dialogue in the format "who says: text"> Avoid: <things you don't want to see, without "no"> Prompt: <the parts above in one flowing text, with the sound in a separate sentence>
You are a directing assistant. Put the parts below together into one flowing video prompt, in Portuguese, for a video generator with audio. Rules: - start with the camera and the subject, then the action, the setting and the light; - describe the sound in separate sentences, at the end; dialogue in the format "who says: text"; - do not add any character, object or place that is not in the parts; - also return the "Avoid" list with nouns only, without the word "no". Parts: <paste here the lines Subject, Action, Setting, Camera, Light and Sound>
Paste the prompt.txt (or upload the file). The page checks the eight lines and whether the text has a shot size and a movement using the terms from lessons 2 and 3 of course 0 (wide shot, close-up, dolly in, static shot, handheld…).
Only for people who already use a terminal. Download the course from GitHub and run it in its folder: the rules are the same as on the page. For your own file, replace --exemplo with its name (prompt.txt); it exits 0 when it passes.
python3 conferir.py prompt --exemplo; echo "saída: $?" python3 conferir.py prompt --errado; echo "saída: $?"
$ python3 conferir.py prompt --exemplo; echo "saída: $?" OK: prompt com 8 partes saída: 0
$ python3 conferir.py prompt --errado; echo "saída: $?" Ainda não. O que falta: - Falta a linha Som: com a descrição (pelo menos 3 letras). - Falta a linha Evitar: com a descrição (pelo menos 3 letras). - Falta citar um movimento de câmera. saída: 1
A shot from your shot list, preferably one with a striking sound.
Pick a shot from your shot list. Fill in the prompt.txt using the template: camera copied from the shot list, light copied from the look card, sound in a separate sentence. If you like, ask the AI assistant to put the flowing text together, and review it. Paste it into "Check your prompt" and download the file to the project folder.
Done when: the page answers OK: prompt with 8 parts and the prompt.txt is in the project folder.
Read the Avoid line and the Prompt line. Replace every sentence with "no" or "don't" ("don't show text", "no people") with things listed out ("on-screen text, people in the background") and check again.
Done when: the page keeps answering OK: prompt with 8 parts and the words "no" and "don't" do not appear in the Avoid line.
Four questions. The explanation shows up after each answer.
Question 1 of 4
Which action leaves the least for the generator to invent?
Question 2 of 4
How does Google's official guide recommend putting sound in the prompt?
Question 3 of 4
You don't want any text on screen. What do you write in the negative prompt field?
Question 4 of 4
Where do the camera and light in a shot's prompt come from?
Pick an answer. The explanation appears after you try.