quest: composable scenes replace canvases; mixes become audio scenes

quest/m0/README.md

@@ -5,7 +5,7 @@
 A native, cross-platform streaming studio (`rs/moq-studio`) for creators and
 teams who rely on OBS today, plus the same app built to wasm. It composites
 cameras, screens, media, browser sources, and remote MoQ guests into scenes
-on the GPU with moq-video and moq-audio, across several canvases at once
+on the GPU with moq-video and moq-audio, publishing several scenes at once
 (landscape and portrait), and goes live to a MoQ relay and directly to RTMP,
 RTMPS, and WHIP ingests.
 
@@ -37,7 +37,7 @@ Decided 2026-10-22 with the maintainer (/quest-plan):
   destination itself. It remuxes its own broadcast through a UI-free egress
   library that a hosted service can also run server-side. There is no
   gateway service and no moq-cli dependency, because a studio user does not
-  run a server. Destinations that take the same canvas and audio mix with
+  run a server. Destinations that take the same picture scene and audio scene with
   identical encode settings share one encoder. The studio picks codecs per destination: Opus by default, AAC
   where an ingest requires it, from platform encoders only.
 - **Extensibility:** first-party features plus browser sources (CEF
@@ -93,13 +93,13 @@ repository.
 - [Wayland display and window sources](/quest/m0/wayland-sources.md) - portal-picked sources that reopen without the picker
 - [ML video effects](/quest/m0/effects.md) - background removal and other ML effects, baked into their source's track
 - [Destinations](/quest/m0/destinations.md) - RTMP, RTMPS, and WHIP push with per-destination codecs and shared encoders, on a UI-free egress library a hosted service can reuse
-- [Multiple canvases](/quest/m0/multi-canvas.md) - landscape and portrait canvases composite at once from shared sources
+- [Composable scenes](/quest/m0/scenes/README.md) - scenes nest scenes and the program, any scene publishes at its own size, and mixes become audio scenes
 - [The join page is live](/quest/m0/join-page-live.md) - `join.studio.moq.dev` serves the join page `--join-page` defaults to
 - [Clap-to-measure sync delay](/quest/m0/sync-clap.md) - a clap sets a microphone's delay against its camera
 - [Audio filters](/quest/m0/audio-filters.md) - per-source noise suppression, gate, compressor, limiter, and plugins, configured from the studio
 - [Browser sources on macOS and Windows](/quest/m0/browser-source.md) - the out-of-process CEF pages Linux has, with shared GPU textures on the other desktops
 - [Web studio](/quest/m0/web.md) - the full studio in wasm on WebGPU and WebCodecs, served from this repository
-- [Virtual camera](/quest/m0/vcam/README.md) - a canvas's program exposed as a system camera on Linux, Windows, and macOS
+- [Virtual camera](/quest/m0/vcam/README.md) - the program or a published scene exposed as a system camera on Linux, Windows, and macOS
 - [Browser sources without an X display](/quest/m0/cef-headless.md) - CEF runs headless, so Wayland sessions and tests need no X server
 - [Signing identities are available](/quest/m0/signing.md) - the Apple and Windows code-signing credentials packaging and the macOS camera need
 - [Packaging](/quest/m0/packaging.md) - signed installers for macOS and Windows and a Flatpak for Linux

quest/m0/destinations.md

@@ -4,7 +4,7 @@
 
 A creator adds any number of RTMP, RTMPS, and WHIP destinations alongside
 MoQ, each with its own codec and bitrate, and goes live to all of them at
-once. Destinations that take the same canvas and audio mix with identical
+once. Destinations that take the same picture scene and audio scene with identical
 encode settings share one encoder. Each destination's status and errors show
 in the UI.
 
@@ -54,7 +54,7 @@ Decided 2026-10-22 with the maintainer:
 - **Benchmarks:**
   - Sweep concurrent broadcasts against destinations per broadcast in the
     library.
-  - Sweep distinct encoder-sharing keys (canvas, mix, and settings) against
+  - Sweep distinct encoder-sharing keys (picture scene, audio scene, and settings) against
     destinations per key in the studio. Finding a destination's encoder must
     not scan every encoder group, and identical keys must never spawn a
     duplicate encoder.

quest/m0/scenes/README.md

@@ -0,0 +1,56 @@
+# Composable scenes
+
+## Goal
+
+Scenes form a graph. A scene can draw another scene, or the live program, as
+a source. Any scene can be published as its own broadcast at its own size, and
+the old audio mixes are audio-only scenes that nest the same way. This
+replaces separate canvases: a portrait broadcast is a published 1080x1920
+scene that draws the program, cropped, with its own overlays.
+
+This epic's own work is an end-to-end test: one studio publishes the program
+at `<--broadcast>` and a portrait scene drawing it at
+`<--broadcast>/<scene>`, each carrying its picked audio scene, and a viewer of
+each sees and hears the same cut at the same moment.
+
+## Plan
+
+Decided 2026-11-05 with the maintainer (/quest-plan), replacing the canvases
+of [moq-dev/studio#42](https://github.com/moq-dev/studio/pull/42) and closing
+[moq-dev/studio#54](https://github.com/moq-dev/studio/pull/54):
+
+- ✅ **Published scenes + scene source** rather than canvases with a layout
+  per scene and linked or unlinked switching. One concept (a scene) instead of
+  two, and no unlinking. Rejected: published scenes that never follow cuts;
+  keeping canvases.
+- ✅ **A scene source draws any scene, or `program`** (whatever is on air,
+  transitions included), scaled and cropped like any picture. Rejected: a
+  program-only source.
+- ✅ **The program stays one switcher at `<--broadcast>`**: Preview, Program
+  and Cut work as today. Other published scenes are always live at
+  `<--broadcast>/<scene>` and follow cuts only through a scene source.
+  Rejected: every published scene switchable on its own.
+- ✅ **Every scene has a size**, defaulting to the program's. The switcher
+  offers only scenes of the program's size.
+- ✅ **Mixes become audio scenes** that hold only sounds and nest, so a
+  "Mics" scene can feed "Stream" and "Headphones". Audio stays independent of
+  cuts, as today. Each published output pairs a picture scene with an audio
+  scene, defaulting to the one on program. Rejected: scenes carrying their own
+  audio, so a cut switches audio; keeping `[mixes]` separate.
+- ✅ **UI:** the Scenes panel lists picture and audio scenes, tagged by kind.
+  The Audio panel stays a mixer (fader, pan, meter, and sends to audio scenes)
+  for live work, as pro switchers do. A graph view of the same data is
+  [its own quest](/quest/m1/scene-graph.md).
+- Nothing is released, so the scene file breaks freely: `[canvases]` and
+  `[mixes]` go away, with no migration.
+
+Carried over from #54: Health grades each published broadcast; a recording
+picks a picture scene and an audio scene; a scene no panel shows and no output
+reads skips drawing. Per-scene cost grows only with what that scene draws;
+`compositor::bench` measures it.
+
+## Required
+
+- [Scene source](/quest/m0/scenes/scene-source.md) - a scene draws another scene, or the live program, as a source
+- [Published scenes](/quest/m0/scenes/published.md) - any scene publishes as its own broadcast at its own size, replacing canvases
+- [Audio scenes](/quest/m0/scenes/audio.md) - mixes become audio-only scenes that nest

quest/m0/scenes/audio.md

@@ -0,0 +1,29 @@
+# [M] Audio scenes
+
+## Goal
+
+The audio mixes become audio-only scenes. An audio scene holds sounds
+(sources' audio, with gain, pan, mute, and delay) and other audio scenes, so a
+"Mics" scene can feed both "Stream" and "Headphones". **On program** picks the
+audio scene the broadcast carries; each published scene and each recording
+picks one. Audio stays independent of cuts.
+
+## Plan
+
+Decided 2026-11-05 with the maintainer (/quest-plan); see the
+[epic](/quest/m0/scenes/README.md).
+
+- Replaces `[mixes]` and `[audio.<source>]` in the scene file with no
+  migration; nothing is released.
+- An audio scene cannot contain itself, directly or through others.
+- The Audio panel stays a mixer: a strip per sound with fader, pan, meter,
+  and a send to each audio scene. Listen and On program work as today.
+- The guest return feed (a mix-minus) and the sync delay keep working; a
+  guest hears the audio scene on program without their own sound.
+- Switching or editing an audio scene never drops or repeats audio, as with
+  mixes today.
+- `README.md` documents audio scenes.
+
+## Required
+
+- [Scene source](/quest/m0/scenes/scene-source.md) - shares the scene graph and its cycle check

quest/m0/scenes/published.md

@@ -0,0 +1,32 @@
+# [L] Published scenes
+
+## Goal
+
+Any scene can be published: ticked **Publish**, it goes live as its own
+broadcast at `<--broadcast>/<scene>`, at its own size, with the audio scene it
+picks (the one on program by default). This replaces canvases: further
+canvases, their per-scene layouts, and the canvas picker are removed, and a
+portrait broadcast is a published portrait scene that draws the program
+through a [scene source](/quest/m0/scenes/scene-source.md).
+
+## Plan
+
+Decided 2026-11-05 with the maintainer (/quest-plan); see the
+[epic](/quest/m0/scenes/README.md).
+
+- Every scene has a size, defaulting to the program's; the switcher offers
+  only scenes of the program's size.
+- The program keeps one switcher at `<--broadcast>`. A published scene is
+  always live and never switched.
+- Each published scene gets the program's two sizes (full and half) and the
+  bandwidth grant shares across all of them, as further canvases did.
+- Carried from [moq-dev/studio#54](https://github.com/moq-dev/studio/pull/54):
+  Health grades each published broadcast with its viewer feedback; a
+  recording picks a picture scene and an audio scene; a scene no panel shows
+  and no output reads skips drawing.
+- Remove `[canvases]` from the scene file with no migration; nothing is
+  released. `README.md` documents published scenes.
+
+## Required
+
+- [Scene source](/quest/m0/scenes/scene-source.md) - a portrait scene follows the program through it

quest/m0/scenes/scene-source.md

@@ -0,0 +1,23 @@
+# [M] Scene source
+
+## Goal
+
+A new source kind, `scene`, draws another scene, or `program` (whatever is on
+air, including a transition in progress), as a picture that an item places,
+scales, crops, and filters like any other. A scene cannot contain itself,
+directly or through other scenes; such an edit is refused with an error.
+
+## Plan
+
+Decided 2026-11-05 with the maintainer (/quest-plan); see the
+[epic](/quest/m0/scenes/README.md) for the model.
+
+- A nested scene renders at its own size into a texture the compositor
+  samples, once per frame however many items draw it, and only while
+  something that is drawn reads it.
+- `program` reads the frame the program output draws, so a scene drawing it
+  shows the same cut on the same frame, with no extra composite.
+- Keyframes, stingers, and restarts in a nested scene start when the scene
+  that draws it goes on program, as its items would.
+- Measure frame time and GPU copies with and without a nested scene.
+- `README.md` documents the source kind.

quest/m0/vcam/README.md

@@ -2,9 +2,9 @@
 
 ## Goal
 
-The studio exposes a canvas's program as a system camera, so video-call and
+The studio exposes the program or a published scene as a system camera, so video-call and
 conferencing apps, which offer no ingest API, can use it. The studio has one
-toggle per canvas, and each OS backend lands on its own.
+toggle per published scene, and each OS backend lands on its own.
 
 ## Plan
 

quest/m0/vcam/linux.md

@@ -2,7 +2,7 @@
 
 ## Goal
 
-The studio publishes a canvas's program as a PipeWire video node that
+The studio publishes the program or a published scene as a PipeWire video node that
 PipeWire-aware apps (browsers and conferencing apps) list as a camera. It
 falls back to an existing v4l2loopback device when one is present.
 The studio's Devices panel leaves out its own virtual camera, so picking it

quest/m0/vcam/macos.md

@@ -3,7 +3,7 @@
 ## Goal
 
 The macOS app bundle carries a CoreMediaIO Camera Extension. Once the user
-approves it, apps list a camera that carries a canvas's program. An app
+approves it, apps list a camera that carries the program or a published scene. An app
 update installs the new extension without a reboot.
 
 ## Required

quest/m0/vcam/windows.md

@@ -2,7 +2,7 @@
 
 ## Goal
 
-On Windows 11, the studio registers a canvas's program as a session-lifetime
+On Windows 11, the studio registers the program or a published scene as a session-lifetime
 camera through `MFCreateVirtualCamera`, backed by a software camera source.
 Apps list it as a camera while the studio runs.
 

quest/m1/README.md

@@ -28,3 +28,4 @@ than opening an m2 for them.
 - [Overlay assets](/quest/m1/overlay-assets.md) - overlays use fonts and images from data: URLs or files beside the scene
 - [Hotkeys and editor preferences](/quest/m1/hotkeys.md) - system-wide hotkeys, arrow nudge, aspect lock, and persisted editor settings
 - [Isolated guest recordings](/quest/m1/guest-recording.md) - record-each-source also writes each guest's tracks to its own file
+- [Scene graph view](/quest/m1/scene-graph.md) - a node view of picture and audio scenes, their routing, volume, pan, and waveforms

quest/m1/scene-graph.md

@@ -0,0 +1,19 @@
+# [M] Scene graph view
+
+## Goal
+
+A Graph panel shows the scene graph: picture and audio scenes as nodes, the
+sources and scenes each one draws or sends to as edges, with each sound's
+volume, pan, and a live waveform on its node. Edges and node settings can be
+edited there, as in the Scenes and Audio panels.
+
+## Plan
+
+Decided 2026-11-05 with the maintainer (/quest-plan): the live controls stay
+a mixer, as pro switchers keep them, and this view is for building and
+inspecting routing, as node tools (TouchDesigner, Fusion) do. It reads and
+edits the same data, so it adds no second model.
+
+## Required
+
+- [Composable scenes](/quest/m0/scenes/README.md) - the graph this view shows