Export: duration sources, audio length, scene/BPM metadata, clean cancel (#523, #524, #528, #532)

[CaYatur]

Closes #523, #524, #528 and #532.

Duration (#523)

• Sources. Length of the audio file (the default, unchanged), a fixed number of seconds, the timeline's length (latest clip, marker or loop end), or N bars. Bars use a typed BPM, or the timeline's tempo map when the BPM is 0.
• Where it is resolved. The main process measures the audio with ffmpeg and resolves the length before starting. The renderer draws ceil(seconds × fps) frames, and frames past the end of the audio are drawn with silence.
• Empty timeline. It is refused with a reason.

Audio follows the video (#524)

• Shorter audio is padded with apad and re-encoded, because a copied stream cannot be filtered.
• Longer audio is cut with -t. MP3 and AAC sources are still copied.
• Per container (from Export: format matrix, alpha and frame sequences (#521, #525, #529) #756): AAC or a copy in MP4, PCM s24 in MOV, Opus in WebM, no audio for GIF, APNG and sequences.

Metadata (#528)

• File name. The save dialog suggests "Track - Scene - 128 BPM". Characters Windows forbids, reserved names (CON, NUL, COM1…) and trailing dots or spaces are removed, and the name is capped at 150 characters.
• Inside the file. comment carries Scene: … · BPM: … next to title, artist and album. MP4 and MOV do not write custom keys, and comment is read everywhere.
• MCP. sceneName is optional and defaults to the active scene. duration is a new argument.

No partial file at the final path (#532)

• Write, then rename. Output goes to name.partial.ext (folder.partial for a sequence) and is renamed only after ffmpeg exits 0. If the rename fails, the job reports an error and names the partial file instead of losing it.
• Cancel or error. The partial output is removed. A renderer crash still keeps what was written, as before.
• Quit during an export. ffmpeg is now stopped and the partial file removed. Before, shutdownCleanup did not touch the export, so ffmpeg could keep running after the app closed.
• Status line. It shows fps and time left, e.g. Rendering [NVIDIA NVENC]… 25% (300 / 1200 frames) · 112 fps · 1:05 left in English. The clock starts at the first frame.

Measured in the app

5 s of audio, isolated profile, run through --bench-export:

job	frames	video / audio	notes
8 s fixed	240	8.000 / 8.000 s	AAC; silence after 5 s (−inf dB); comment Scene: Kulüp
3 s, WAV	90	3.000 / 3.000 s	AAC
3 s, MP3	90	3.000 / 3.005 s	copied; packet boundary
8 bars @ 120	480	16.000 / 16.000 s	comment BPM: 120
timeline (marker at 6 s)	180	6.000 / 6.000 s
empty timeline	—	refused	"The timeline is empty, so it cannot set the duration."
cancel after 2.5 s	—	—	no file, no .partial
quit after 3 s	—	—	no file, no .partial, no ffmpeg process left

A 6 s PNG sequence over 5 s of audio wrote 180 frames into the final folder.

Verification

• npm test: 3132/3132.
  • tests/export-job.test.js covers duration resolution, the pad/trim/copy rules, metadata, file names, the partial path, ETA, and the main, panel and MCP wiring.
  • tests/export-quit.test.js now checks the partial and promote path, and that quitting stops the export.
• --smoke --smoke-export with an isolated profile: PASS, with no .partial left behind.
• Panel over CDP:
  • Choosing "Sabit süre" shows the seconds slider (0:30), and choosing bars shows Bars and BPM.
  • The English labels and the status line translate.

[cayadraft]

@CaYatur

CaYaDraft — pull request

Summary
Export: duration sources, audio length, scene/BPM metadata, clean cancel (#523, #524, #528, #532)

Files
CHANGELOG.md (modified, +13/-0)
@@ -36,6 +36,19 @@ GitHub. The full reasoning and the measurements behind each change are in [ROADM
31 s to 11 s, MilkDrop at 1080p60 from 31 s to 13 s, and MilkDrop at 4K from 60–67 s to 18–19 s.
The conversion matches ffmpeg's to within one level on luma; the scene itself is drawn exactly as
before.
+- Choose how long the export runs [523]. The length of the audio file (as before),…

README.md (modified, +5/-3)
@@ -17,7 +17,7 @@ layers and 40 GPU effects, an OBS overlay, Spout / Syphon, **project


[![Platform](https://img.shields.io/badge/platform-Windows%20%…

README.tr.md (modified, +5/-3)
@@ -18,7 +18,7 @@ destesi**, ve MCP üzerinden yapay zekâ kontrolü.


[![Platform](https://img.shields.io/badge/platform-Windows%20%7C%20macOS%20%7C%20Linux-111997.sv…

ROADMAP.md (modified, +31/-2)
@@ -93,7 +93,7 @@ hands, and a row only moves up once it ships.
|---|:--:|---|
| Per-app audio capture on macOS and Linux | v3.1.6 | On main (#514, #517). Linux verified in WSLg with PulseAudio; macOS written on ScreenCaptureKit and compiled in CI, never run on a Mac. Windows shipped in v3.1.3 |
| Live level meter per application | v3.1.6 | Done on main (#516): a moving dB-scale bar per ap…

src/admin/admin.js (modified, +62/-3)
@@ -3385,6 +3385,30 @@
type: 'select', path: 'export.format', label: 'Biçim', rebuild: true,
options: () => exportFormatOptions(),
},

•      {
  

•        type: 'select', path: 'export.duration.mode', label: 'Süre', rebuild: true,
  

•        options: [
  

•          { value: 'audio', label: 'Ses dosyasının süresi' },
  

•          { value: 'seconds',…
  

src/exporter/exporter.js (modified, +6/-1)
@@ -423,7 +423,12 @@
if (prof) { prof.decodeMs = performance.now() - tp; tp = performance.now(); }
await setup(job);
if (prof) prof.setupMs = performance.now() - tp;

• totalFrames = Math.max(1, Math.ceil(duration * fps));

• /* Süre ana süreçten gelebilir (Export: duration from several sources #523): sabi…

Note
The local model is unavailable, so this is only the title and file list.

---

Stopping here. I will not merge, push, or change this pull request until you ask.

/caya explain describes the diff. /caya plan previews the next edit. /caya revise adds a commit.