Lesson 2 · 18 min
In 2026 sound stopped being the last step: the main generators return ambience, effects and speech together with the image. This forces you to decide, shot by shot, what is born with the video and what comes in later, like the score and the narration. This lesson settles that decision in a sound map.
Start the lesson
Goal: decide, shot by shot, what is heard and what comes out with the video or separately, and build the scene's sound map.
Five layers. Click each one and see where it usually comes from.
Rule of thumb: what comes from the action of the shot can come out with the video; what crosses shots goes into the edit.
In the scene "The first batch", using the sound column of the shot list from course 0.
Native audio became the standard in 2026 in the main generators (dated examples, Oct/2026: Veo 3.1, Kling 3.0, Seedance 2.x, Runway Gen-4.5). Seedance 2.5 accepts up to 10 reference audios per generation.
Use native for what comes from the shot: ambience, the effect of the action, a short line. Keep separate what crosses shots, what has to sound the same in all of them, and what will change language.
Wind and knocks on the door: native. Score and opening narration: separate, in the edit.
Diegetic sound exists in the scene and the character hears it; non-diegetic sound only the audience hears, like the score. A well-resolved shot has ambience and the effect of the action.
Silence is a choice too: cut the music before the turning point, leave one shot with only the ambience, a breath after the impact. If you can show it, don't narrate it.
Music generators change fast: Suno released v5.5 on 03/26/2026; a v6 and ElevenLabs' Eleven v3 appear only as reported, not confirmed by an official source. Read each tool's license of use before you publish.
Synthetic or cloned voice: only your own, or from someone who authorized it in writing (lesson 7). Free, local route: Concat cleans and levels the voice and has an AI voice.
The voice dominates; the music stays atmospheric, underneath; the effect appears at the moment of the action. The most common mistake is leaving the music at the same volume as the voice.
Starting point: lower the score until the line is clear without effort, then check with headphones. Leave a few extra seconds of score for the exit fade.
The format the page checks, and a request for the AI assistant to draft the map from the shot list.
Scene: <name> (<duration> s) Score: <style, where it comes in and where it goes out; with or without vocals> # n | what you hear | origin 1 | ambience: <...>; effect: <...> | origin: <native or separate> 2 | line: "<short sentence>" | origin: <native or separate>
You are a film sound editor. Read the shot list below and propose a sound map, one line per shot, in this format: n | what you hear (ambience, effect, line, score or silence) | origin: native or separate Rules: - native = comes out with the generated video (ambience, effect of the action, short line); - separate = goes in the edit (score, long narration, sound that repeats the same across several shots); - only one origin per shot; - leave at least one shot with only ambience before a line; - on the first line, describe the score: style, where it comes in and where it goes out, no vocals under the line. Shot list: <paste your shot list here>
Paste the file (or upload the .txt). Test first with the example that passes and the example with errors.
Only for people who already use a terminal. In the folder of the course downloaded from GitHub; for your own file, replace --exemplo with its name (mapa-som.txt).
python3 conferir.py som --exemplo; echo "saída: $?" python3 conferir.py som --errado; echo "saída: $?"
$ python3 conferir.py som --exemplo; echo "saída: $?" OK: 6 planos com camada de som e origem decidida saída: 0
$ python3 conferir.py som --errado; echo "saída: $?" Ainda não. O que falta: - Falta a linha Trilha: com a descrição (pelo menos 8 letras). - Plano 2: 2 escolhas de origem; fique com uma só. - Plano 3: falta camada de som. - Plano 4: falta camada de som. - Plano 4: falta origem. saída: 1
The sound map of your scene, built from the sound column of your shot list.
Write the Trilha line and one line per shot, with what you hear and the origin (nativo or separado). Paste it into "Check your sound map" and download the mapa-som.txt to the project folder.
Done when: the page answers OK: [n] planos com camada de som e origem decidida and the mapa-som.txt file is in the project folder.
Pick the shot before the turning point of your scene and leave only ambience or silence in it, with no score and no line. Check again.
Done when: the page keeps answering OK: … and the shot before the turning point has only ambience or silence in your list.
Four questions. The explanation appears after each answer.
Question 1 of 4
The score of your film runs across all six shots. Where should it come from?
Question 2 of 4
Which of these layers usually comes out well as native sound, together with the video?
Question 3 of 4
Your mix is ready and the line disappears when the score rises. What should you fix?
Question 4 of 4
You want to use a cloned voice in the narration. Which voice can it be?
Choose an answer. The explanation appears after you try.