A sound library you can play by ear
War Room 02 turns a small set of generated sounds into an editable cinematic scene. The sound direction is deep resonant brass, physical percussion, soaring strings and heroic scale. These are original generated sounds; no soundtrack recordings were sampled.
What is ready
Twelve sounds: four isolated percussion impacts, two low textures, two brass phrases, three high string textures or phrases and one airy harmonic texture. Three starting scenes combine them: Gathering storm, Battle charge and The hero rises.
Five layers make a scene: pulse, weight, movement, statement and atmosphere. Choose a sound, place its entrance and change its loudness. Edit the drum phrase by clicking squares. Pace, intensity, room and movement controls let you explore without reading notation. Previewing a library sound pauses the scene so it can be heard alone.
Each scene lasts four drum rounds and repeats. Musical layers restart each scene at their chosen entrance and fade at its boundary. Later entrances intentionally play only the beginning of a phrase. Save audio renders two scenes plus a four-second tail. Save sketch preserves the settings as JSON. The browser also remembers the last sketch locally; it is not an account or cloud backup. Backgrounding the tab pauses playback.
The important separation
Create sounds once → store the library → arrange reusable sounds → hear changes locally → save audio or an editable sketch.
Current
All twelve sounds are pre-generated. Eight musical layers use Eleven Music v2.5; four percussion impacts use Sound Effects v2. The site includes six prepared pace versions for each musical layer, from 72 to 132 beats per minute. FFmpeg time stretching changes duration without intentionally transposing the sound. Selecting a new pace or changing a sound or entrance restarts the scene. Loudness, intensity, room and drum edits update during playback.
Proposed
Keep this interaction as the foundation, then add longer sections, moveable phrase blocks, phrase trimming and custom sound imports. For production, replace generated layers with recorded or licensed orchestral instruments. Pitched note editing and expressive articulations need a separate instrument layer; they cannot be recovered simply by stretching these audio phrases.
Why these tools
| Tool | What was verified | Decision |
|---|---|---|
| Eleven Music v2.5 | The current API documentation lists this model; actual generation succeeded with the available connection. Eight ten-second orchestral clips were created. | Used for brass and strings because these need musical tone and phrasing. |
| Eleven Sound Effects v2 | Actual generation succeeded for four isolated impacts. | Used for percussion one-shots; the rhythm is controlled locally. |
| Stable Audio 3.0 | Official documentation describes text-to-audio, audio-to-audio and inpainting, with six-minute compositions. No comparison render was made here. | A candidate for the next controlled sound-quality comparison, especially transforming an existing sample. Not established as better. |
Sources checked 26 September 2026: ElevenLabs Music API, ElevenLabs Sound Effects, Stable Audio 3.0, Stability API documentation.
Sound preparation and limits
Original MP3 generations were filtered by instrument range, peak-adjusted and given short edge fades. Percussion received mild noise reduction and leading-silence trimming where detected. Sustained textures are extended from stable middle sections with overlapping fades to reduce premature dropouts. Musical layers use prepared tempo versions so strings do not simply rise in pitch when sped up. The waveform thumbnails are calculated from the audio.
Musical prompts requested a common D-minor center and 96 BPM for rhythmic phrases. Generative output does not guarantee exact tuning, stem isolation or rhythmic alignment. Sustained textures tolerate layering best; moving phrases can drift or clash and should be auditioned together. The model can add sounds beyond the requested instrument. These are prompted instrument layers, not stems separated from one performance. This is a creative prototype, not a verified orchestral sample library or mastered score.
No generation API is called by the browser, and no credentials ship with the site. Sound files are downloaded on demand and cached in memory for the session; normal browser HTTP caching can also reuse files. Downloading the sound pack provides the twelve prepared original-pace MP3s for local use.
The temporary site is a separate Cloudflare Pages project. It does not change an existing client site, domain or DNS record. Cloudflare Pages URLs do not automatically expire; this prototype remains online until the project is removed.