The existing study clips are silent. These notes come from the original soundtracks: opening, middle and closing excerpts, 24 seconds each. Machine analysis, not human listening verification.
Texture reference
Introducing ChatGPT Atlas
The analysis describes an electronic pulse, rounded low tones, light bell- or mallet-like plucks and soft sustained layers. It reports restrained changes in layering rather than big builds. Exact instruments are uncertain.
Proposed use. Try a few soft plucks over a low pulse. Thin the percussion when speech needs more room.
All three samples were classified as speech over a musical bed. The analysis describes a steady, moderate pulse, soft pad- or keyboard-like sounds and light percussion, with small arrangement changes rather than dramatic drops.
Proposed use. Keep a consistent groove through the demo. Use small changes in texture instead of chasing every UI action with a musical hit.
The model reported no detectable music in the opening and closing samples, but a subdued bed in the middle sample. Treat the absence judgments as provisional: quiet music can be missed. This is not evidence of one continuous soundtrack style across all three films.
Proposed use. Leave room for voice-only passages. A dense explanation does not need continuous music.
What this supports. A sparse pulse-and-pluck direction to try, not a single “OpenAI sound.” The study covers 216 seconds, not all 100 films or the complete three soundtracks. Instruments are tentative; BPM, key, mix levels and exact musical transition times were not measured. Links open the originals near each sample’s start; they do not isolate or stop at the end of the window.
Method and provenance
Machine audio analysis of nine 24-second excerpts from three original source soundtracks (216 seconds total). No human listening verification or complete soundtrack review. The existing visual-study clips are silent and were not used to infer music.
Public source audio was decoded into stereo MP3 analysis excerpts. The audio-understanding model was gpt-audio-2025-08-28. Sample positions and content hashes are in the study data. No source soundtrack, transcript or extracted audio is republished here. Exact source credits and low-level music-presence judgments remain unverified.
02 / The prompt
Start with this cue.
Written for a fresh instrumental composition. It borrows a sparse arrangement approach, not melodies, branding or recordings. The 90-second shape is for auditioning; it is not a presentation timing plan.
ElevenLabs / Ready to paste
Create an original 90-second instrumental background cue for a spoken product presentation with animated software demos.
Minimal electronic music with a warm, curious, quietly confident mood. Aim for approximately 104 BPM in a steady four-beat pulse. Use a rounded, restrained synth bass, soft mallet-like plucks, a little airy sustained harmony, and very light dry electronic percussion. Keep the arrangement sparse and the melody simple: a short original motif with plenty of space between notes. Let the low pulse provide forward motion without a big kick drum. Keep bright percussion and busy midrange patterns out of the narrator's way.
Suggested shape: 0–8 seconds, introduce a soft texture and a few isolated notes; 8–32 seconds, settle into a gentle pulse; 32–64 seconds, sustain the groove with small additions and subtractions; 64–80 seconds, add a modest lift in warmth, not loudness or density; 80–90 seconds, remove percussion and resolve with a short, soft decay to silence. Keep section changes smooth, with clean places to edit between phrases. These are musical sections, not cues synchronized to a finished film.
The music should remain interesting at low volume while spoken narration stays in front. No vocals, lyrics, spoken words, choir or vocal chops. No dramatic risers, cinematic impacts, big drops, cymbal crashes, heroic orchestral swells, busy solos, notification sounds or mouse-click effects. Avoid a triumphant advertising jingle. Compose a new melody and arrangement rather than reproducing an existing soundtrack.
If it feels too busy, ask for fewer plucks and no upper percussion. If it feels static, ask for a slightly clearer low pulse before adding more instruments. Audition the result; a prompt does not guarantee exact section timing.
03 / Generate when ready
One request. No key in the page.
Download the Python helper and request JSON into the same folder, then run the command there. A real run uses your ElevenLabs credits; this page never submits a generation request.
The command is with-secrets ai-dev <command>. The helper reads ELEVENLABS_API_KEY from the environment; it never puts the key in browser code or command-line arguments. It refuses to overwrite an existing output and rejects error responses rather than saving them as music.
Request settings and a no-cost dry run
The request uses model_id: music_v2_5, music_length_ms: 90000 and force_instrumental: true. It calls the music endpoint, not text-to-speech or sound effects. The output is explicitly requested as mp3_44100_128 for auditioning. Account access is not tested by this page.
The dry run validates and prints the prompt payload without a key, network request or generation charge. To change duration, update both the JSON length and the prompt’s arrangement. The API documents a 3–600 second range; prompts and composition plans are alternative input modes.
API documentation checked September 18, 2026. Models and account access can change.
This prompt does not approve a presentation render or set its playback offsets. Choose the actual music placement with the storyboard and the supplied recording.
Audition with the actual voice
Check a dense explanation and a quiet pause through the intended meeting setup. Lower the music if any words become harder to follow. Do not pick a final level from the generated file alone.
Keep the two videos separate
The talking head remains the OBS asset. If the music is approved, route it once with the presentation; do not duplicate the narration or mix in the reference-film audio.
Use the visible start cue
Keep the presentation’s opening lead-in and the recognizable content transition that cues the OBS switch. An 8-second musical intro is not a measured playback offset or a replacement for that cue.
Finish before the ending buffer
Trim or fade the approved score with the presentation, before the talking head’s wave and pause end. Set that point on the shared timeline from the real footage, not from this 90-second audition.
Before using a take: listen for unwanted voices, distracting motifs, abrupt endings and section drift; check it against the real narration; confirm the account’s applicable usage terms. Keep the generated music file separate until the score and timing are approved.