THE LANTERN STAYS — TECHNICAL PRODUCTION RECORD

GENERATION SETTINGS (verified inputs, not a transcription)
Engine: ACE-Step 1.5.0, official source commit ca1e85fe9430179831e6bc6be790c332190a3866.
Checkpoint: ACE-Step/Ace-Step1.5, revision 19671f406d603126926c1b7e2adc169acbcade22, acestep-v15-turbo.
Renderer: CPU PyTorch 2.10.0+cpu; DiT/text encoding BF16 through a local opt-in wrapper, VAE upstream CPU float32; SDPA; no GPU; no external generation API.
Mode: text-to-music, lyrics supplied, no LM enhancement, 8 diffusion steps, Euler ODE, seed 20261106, one output.
Requested duration: 60 seconds. Requested tempo: 84 BPM. Requested meter: 4/4. Requested key: G major. Language: English.
Requested sound: intimate acoustic folk, a gentle sung voice, fingerpicked steel-string guitar, upright bass, brushed drums, sparse piano answers, clear diction, patient and hopeful mood, organic phrasing, quiet resolved ending. No named artist or borrowed melody.

COMPOSITION INTENT (not verified rendered notes or chords)
The first four lines establish rain, patience and an opening door; the next four turn an empty room into movement; the closing four return to the lantern and resolve toward warmth.
A suitable arranged score could use G–D–Em–C, then Em–C–G–D, ending C–D–G. These chords were NOT provided as event-level constraints and are NOT a transcription of the generated audio.
At 84 BPM a quarter note lasts 0.714286 seconds; a 4/4 bar lasts 2.857143 seconds. Sixty seconds spans 21 nominal bars. The generator may vary the actual tempo and timing.

LIMITS OF THIS RECORD
ACE-Step produces an audio waveform, not an authoritative score or multitrack session. Exact sung pitches, chord voicings, individual instrument notes, articulations and note onsets have not been transcribed or verified. No MIDI, stems, exact SongIR lock or full note-for-note record is claimed. Those require a separate transcription/reconstruction and listening review.

LICENSE AND PROVENANCE
ACE-Step code/checkpoint licensing is MIT; Qwen embedding model is Apache-2.0. These are open-source/permissive licenses, not public-domain declarations. Training provenance is mixed, not demonstrated exclusively public domain. No paid generation service or third-party song sample was used as an audio reference in this render. Lyrics were written for this request. Automated generation does not guarantee copyright clearance or worldwide copyright ownership.

VERIFIED OUTPUT CHECKS
{
  "sample_rate": 48000,
  "channels": 2,
  "seconds": 60.0,
  "peak": 0.8912353515625,
  "rms": 0.14161449422079112,
  "finite": true,
  "m4a": {
    "programs": [],
    "stream_groups": [],
    "streams": [
      {
        "codec_name": "aac",
        "sample_rate": "48000",
        "channels": 2
      }
    ],
    "format": {
      "duration": "60.000000"
    }
  },
  "sha256": "18aad600a95db99916e3023907cda63adff8a052458df98b2dc4452cb8c49cba"
}
Render completed in approximately 169.5 seconds. Audio was encoded to AAC in an M4A container. Waveform checks passed: finite values, non-silent, 60 seconds, stereo 48 kHz. No full-audio listening review was performed; sung wording, musical quality and exact notes remain unverified.