Deliver consistent-character, consistent-voice FableLoom production and hosted play

[atomantic]

Goal

Deliver the production architecture described in docs/plans/2026-08-29-fableloom-character-voice-hosted-production.md: repeatable character identity and voice across full movies or episodic FableLoom stories, plus a two-device hosted experience where a viewer speaks to an off-screen protagonist over authored, pre-rendered video.

Why

PortOS already has a Universe Bible, character reference sheets and LoRAs, Kokoro/Piper synthesis, a branching FableLoom runtime, and local STT/TTS transport. The remaining work is to join those primitives into versioned character production packages, explicit playback assets, production provenance, and a safe half-duplex hosted session protocol.

Delivery slices

Child issues will cover:

• approved Universe character production packages;
• visual canon bindings and capability-aware conditioning;
• machine-local voice profiles using existing engines;
• Qwen3-TTS voice design, consented cloning, and optional fine-tuning;
• entry, hold-loop, and transition playback assets;
• scoped QR-hosted sessions with half-duplex character voice;
• episodic production orchestration and continuity review; and
• a non-blocking future investigation of diegetic FaceTime calls.

Program acceptance criteria

• Canon-locked shots identify the exact character, approved identity assets, wardrobe, adapters, temporal source, and omitted inputs.
• Offline dialogue and hosted replies resolve an approved character voice revision and reproducible mastering chain.
• Missing local artifacts degrade visibly instead of silently substituting or appearing intentionally empty.
• Hosted mode requires HTTPS, an off-screen protagonist, safe rendered hold loops, and ready STT/LLM/TTS routes.
• Viewer and protagonist turns are half-duplex, and live character voice never overlaps rendered character dialogue.
• All provider-heavy generation, training, and review remains explicitly user-triggered.

Out of scope

• Real-time AI video or runtime lip-sync.
• Internet relay or public exposure of a PortOS instance in the first hosted release.
• Multi-audience sessions in the first release.
• FaceTime integration as a dependency of browser-hosted mode.

Tracking

• Add approved character production packages to Universe #5378 — approved Universe character production packages
• Add FableLoom visual canon bindings and capability-aware conditioning #5379 — visual canon bindings and capability-aware conditioning
• Add machine-local character voice profiles using Kokoro and Piper #5380 — machine-local Kokoro/Piper character voice profiles
• Add Qwen3-TTS voice design, consented cloning, and optional fine-tuning #5381 — Qwen3-TTS design, consented cloning, and optional fine-tuning
• Add FableLoom entry, hold-loop, and transition playback assets #5382 — entry, hold-loop, and transition playback assets
• Add scoped QR-hosted FableLoom sessions with half-duplex protagonist voice #5383 — scoped QR-hosted sessions and half-duplex protagonist voice
• Add episodic FableLoom production orchestration and continuity review #5384 — episodic production orchestration and continuity review
• Investigate diegetic FaceTime calls as a FableLoom game mechanic #5385 — non-blocking future FaceTime game-mechanic investigation

The browser-hosted path is the delivery target. #5385 is related future work
and does not block this epic's initial hosted release.

[atomantic]

Every delivery slice has shipped, so this epic is complete:

• Add approved character production packages to Universe #5378 — approved Universe character production packages → feat(universe): add approved character production packages #5390
• Add FableLoom visual canon bindings and capability-aware conditioning #5379 — visual canon bindings and capability-aware conditioning → feat: add FableLoom visual canon conditioning #5400
• Add machine-local character voice profiles using Kokoro and Piper #5380 — machine-local Kokoro/Piper character voice profiles → feat(voice): add Qwen3-TTS voice design, consented cloning, and fine-tuning (#5381) #5402
• Add Qwen3-TTS voice design, consented cloning, and optional fine-tuning #5381 — Qwen3-TTS design, consented cloning, and optional fine-tuning → feat(voice): add Qwen3-TTS voice design, consented cloning, and fine-tuning (#5381) #5402
• Add FableLoom entry, hold-loop, and transition playback assets #5382 — entry, hold-loop, and transition playback assets → feat(fableloom): add entry, hold-loop, and transition playback assets (#5382) #5401
• Add scoped QR-hosted FableLoom sessions with half-duplex protagonist voice #5383 — scoped QR-hosted sessions and half-duplex protagonist voice → Add scoped QR-hosted FableLoom sessions with half-duplex protagonist voice (#5383) #5404
• Add episodic FableLoom production orchestration and continuity review #5384 — episodic production orchestration and continuity review → feat(fableloom): orchestrate episodic production and continuity review #5410
• Investigate diegetic FaceTime calls as a FableLoom game mechanic #5385 — FaceTime game-mechanic investigation → docs: decide diegetic FaceTime transport boundaries (#5385) #5414

Closing as completed; the design record stays at docs/plans/2026-08-29-fableloom-character-voice-hosted-production.md.