fix(mix): no tirar el audio cuando solo algunos clips tienen pista

[cursor]

El join suave (freeze-tail + crossfade) sondaba únicamente el primer clip para decidir si el montaje llevaba audio.

• Bumper o plano visual-only al inicio: el mix salía mudo y se perdía el diálogo de los clips siguientes.
• Diálogo al inicio y B-roll sin pista después: ffmpeg pedía [n:a] inexistente, el join fallaba y el fallback duro también, así que el film no se ensamblaba.

Qué cambia

• Se sonda cada clip.
• Si alguno tiene audio, el mix lo conserva.
• Los clips sin pista reciben silencio estéreo 48 kHz (anullsrc) para que acrossfade no rompa.
• Si ninguno tiene audio, el montaje sigue siendo solo vídeo.

Cómo reproducir

1. Ensamblar 3+ clips sin banda sonora externa (Director / Series / rejoin de galería).
2. El primero es vídeo-only (Wan, import, bumper) y los siguientes tienen diálogo H3 → el MP4 final salía silencioso.
3. O al revés: diálogo primero y un B-roll sin audio → el ensamblado fallaba.

Validación
pytest tests/test_mix_concat.py — 9 passed.

  

---

Note

Medium Risk
Changes ffmpeg filter graphs for a core assembly path; incorrect silence/stream handling could still affect dialogue or cause encode failures, though behavior is covered by new unit tests.

Overview
Fixes hold-tail + crossfade assembly when clips mix video-only segments with dialogue: the pipeline used to probe only the first clip for audio, which could mute the whole export or make ffmpeg fail on missing [n:a] streams.

Per-clip probing (probe_has_audio / probe_audio_flags) now decides whether to mix audio if any clip has a track. build_hold_crossfade_filter accepts optional has_audio flags and, for clips without audio, inserts 48 kHz stereo silence via anullsrc (with resample/format on real tracks) so acrossfade stays valid. If no clip has audio, the graph stays video-only as before.

concat_with_tail_hold_and_crossfade wires this in by replacing the first-clip ffprobe with flags for every input. Tests cover silent-first, silent-later, all-silent, and per-path probing.

^{Reviewed by Cursor Bugbot for commit c28e73c. Configure here.}

[github-actions]

PR Review — Loreframe Studio

Risk: low
Scope: 2 file(s); +123/-20; backend services

Automated review from scripts/analyze_pr.py. This is a heuristic pass (no LLM) so humans still own the merge decision.

Findings

• No heuristic issues. Still run the CI checklist below.

Changed files

• modified: app/services/mix_concat.py, tests/test_mix_concat.py

CONTRIBUTING checklist

• python scripts/verify_clean_repo.py
• python -m compileall -q app/services app/launch.py scripts
• cd ui && npm run build if the UI changed
• No weights, CivitAI sidecars, or generated guides
• Stays local-first (no required accounts / telemetry)

Posted by the repo PR review workflow. Re-runs on each push to the PR.

[cursor]

Risk: medium. Approval criteria are met: Cursor Bugbot and Cursor Security Agent completed successfully with no findings that need human review, but GitHub blocked this automation from approving its own PR. No reviewers assigned.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}