GENAI 101 - Module 1: Music

Lesson 1.3: Scoring a Scene with AI-Generated Soundtracks

Estimated time: 90 minutes. Competencies: problem solving, AI feedback evaluation.

Video: Lesson 1.3: Scoring a Scene with AI-Generated Soundtracks. Four-minute introduction: cue sheets, the six evaluation criteria, and the three fixes for a cue that misses. (Instructor: build with make_videos.py or record and embed your own.)

SOUNDRAW: AI Music Composition. Demonstration of mood-based soundtrack generation with Soundraw, one of the tools suggested for this lesson.

Learning objectives

By the end of this lesson you will be able to:

Translate a scene's dramatic beats into a cue sheet with mood, tempo, and instrumentation targets.
Generate instrumental cues with a GenAI tool and evaluate each against the cue sheet using explicit criteria.
Diagnose a mismatched cue and apply one of three fixes: revise the prompt, change the tool, or edit the audio.
Assemble a scored scene of 30 to 60 seconds with two contrasting cues.

Warm-up: what does the music know? (5 minutes)

Watch 30 seconds of any film scene with the sound off. Write down what you think is happening emotionally. Now watch it with the sound on. Did the score confirm your reading, add something, or contradict it? Film composers call the last option "playing against the scene." You will make that choice deliberately today.

Key ideas

Scoring starts with a cue sheet

A cue sheet lists each moment that needs music, when it starts and ends, and what it should do. Before you touch a tool, you decide: does this cue establish a place, reveal a feeling, build tension, or release it? A cue sheet turns a vague request ("make it sad") into a brief the model can act on and a standard you can evaluate against.

Evaluating AI output is a skill, not a reaction

"I like it" is not evaluation. Evaluation compares the output with the brief on named criteria. For a cue, the criteria are:

Criterion	Question
Mood match	Does the emotional color match the cue sheet?
Tempo and energy	Does the pacing match the action on screen?
Instrumentation	Are the sounds appropriate to the setting and period?
Structure	Does the cue start, build, and end where the scene needs it to?
Technical quality	Is it free of glitches, abrupt loops, and muddy mixing?
Fit under dialogue	Will it sit under speech without fighting it?

Rate each from 1 to 4. Anything at 1 or 2 is a problem to solve, not a reason to give up.

Three fixes, in order

When a cue misses, work through these in order:

Revise the prompt. Most misses come from underspecified mood or tempo. Add a reference ("in the style of a slow-burning noir cue"), a tempo, and a shape ("starts sparse, swells at 20 seconds").
Change the tool or mode. If the model keeps adding vocals, switch to instrumental mode. If it cannot do the genre, try Soundraw or Mubert, which are built for mood-based cues.
Edit the audio. Trim, fade, layer, or slow the cue in Soundtrap. Sometimes the best fix is a 12-second crossfade you make yourself.

Diagnosing which fix applies is the problem-solving competency in action.

Activity: score a scene (60 minutes)

Choose a scene. Use a 30 to 60 second clip you have the right to use: a public-domain film clip, your own rehearsal footage without identifiable students, or a silent animatic you build from three still images in Canva.
Write the cue sheet. Identify two contrasting cues. Example: Cue A (0:00 to 0:25) establishes a calm kitchen at dawn; Cue B (0:25 to 0:50) shifts to dread as a letter is opened. For each cue write mood, tempo, instrumentation, and shape.
Generate Cue A. In Suno's instrumental mode (or Soundraw), write a prompt from the cue sheet. Generate two candidates. Score both on the six criteria.
Generate Cue B. Repeat. Make sure Cue B contrasts with Cue A in at least tempo and instrumentation.
Diagnose and fix. Pick the weakest score across your four candidates. Apply one of the three fixes and regenerate or edit. Record the diagnosis and the fix in your notes.
Assemble. In Soundtrap or a video editor (Canva, CapCut, iMovie), place the clip on the timeline, add Cue A and Cue B, and create a transition between them (a crossfade, a hard cut on a visual beat, or a moment of silence).
Export as MP4 and name it Lesson-1-3_LastName_Title.mp4. If you cannot export video, export the two cues as MP3 and submit them with the cue sheet.

Prompt bank for cues

"Instrumental, cinematic ambient, 60 BPM, soft piano and warm string pad, gentle and hopeful, slow swell over 30 seconds"
"Instrumental, tense underscore, 84 BPM, low cello drone, ticking percussion, rising dissonant strings, no melody"
"Instrumental, playful chamber cue, 132 BPM, pizzicato strings and clarinet, light and mischievous, ends on a question"
"Instrumental, West African kora and percussion, 100 BPM, bright and communal, steady groove"

Create and share

Post in the Lesson 1.3 Gallery:

Your scored scene (MP4 or link), or two cues plus the cue sheet.
Your cue sheet.
Your scores for all four candidates on the six criteria.
A rationale of no more than 75 words: the cue you diagnosed, the fix you chose, and why.
Your one-line AI credit.

Reply to two peers: name one cue that works, and one moment where you would have "played against the scene."

Resources

Suno instrumental mode; Soundraw and Mubert as mood-based alternatives.
Public-domain film sources: the Prelinger Archives on archive.org.
Canva, CapCut, or iMovie for assembly.

Reflection prompts

Which criterion did the AI fail most often, and what does that tell you about how it was trained?
When was editing the audio yourself faster than regenerating?
How would you use the six-criteria grid as a rubric for a student scoring project?