Desktop Studio production manual

Install the build intended for your operating system and CPU architecture. Native plugins, device APIs and optional sidecars have platform-specific requirements; a source implementation is not proof of a signed distributable installer. The currently tested macOS paths do not establish Windows/Linux or every vendor plugin as accepted.

Use the app license/trial interface and the entitlement shown by each command. An expired trial can block licensed native operations. A web subscription is not automatically a Desktop serial number. If an existing license is rejected, retain the displayed error and contact support; do not replace the app with an untrusted activation patch.

1. Choose a writable project folder and keep its assets alongside the project or in a collected package.
2. Select your audio output, then test at a low monitoring level.
3. Choose a small built-in pattern or factory score before adding third-party plugins.
4. Save the initial project and reopen it before beginning a long recording.

Optional offline narration requires a separately prepared model/runtime payload. Standard builds do not automatically contain that large payload. Export formats requiring an encoder also require the appropriate shipped or configured encoder executable.

Stop or pause the relevant transport before structural changes requested by a panel. Studio, arrangement and scripting transports are distinct owners; preparation and native output leases can refuse conflicting playback or recording.

.gmstudio stores the versioned DAW arrangement/session, including extensions such as video edits and supported take/routing state. .gms stores composition source. .gmd is a separate Studio project format used by other project workflows. Renaming an extension does not convert the content.

1. Save a named project before a major edit. Confirm the intended file and folder.
2. Use Open project and respond to the unsaved-work prompt.
3. If media moved, use the relevant relink workflow and verify its duration/content rather than editing raw project paths.
4. Use Collect project with media for an independent portable folder. Keep the source project until the collected copy opens successfully.
5. Move the whole collected folder, then reopen its project and inspect media, plugin availability and routing.

Collection is a copy operation. Plugin binaries and licenses are not made portable merely by saving their state. Derived audio can still need originals for Restore or Unfreeze. Back up important projects independently of recovery caches.

1. Create or select audio and MIDI tracks. Import source material or use a built-in instrument.
2. Place regions on the timeline. Trim, split, move and duplicate while preserving an untouched alternative when needed.
3. Edit MIDI notes and their expression lanes. Use the tempo map deliberately: musical ticks and fixed picture times are different coordinates.
4. Route tracks to buses/sends, then check mute/solo and levels before playback.
5. Save a checkpoint before destructive consolidation or flattening.

Undo/redo covers supported arrangement edits. A failed or stale background operation should not silently replace a newer session. If preparation is canceled after changing the project, reselect the intended region and retry from the new state.

Web Audio output and native output use separate device paths. Select the intended route in each applicable control. Native playback/capture owns output leases so incompatible streams cannot quietly compete. A device that appears by name may still lack the required channel count, sample rate or buffer format.

Use interface direct monitoring for the native recording workflows that request it. Do not confuse input monitoring gain with printed source gain or assume recorded audio has already been latency-corrected. Match duplex input/output sample rates and make a short recorded test before an important take.

For latency calibration, use the provided cable measurement workflow with the selected device route. Keep monitoring conservative, run the required probes, and save/apply only a successful matching profile. A profile is bound to its device configuration; changed devices/rates invalidate assumptions. Callback clock alignment and physical round-trip latency are separate measurements.

1. In the recording controls, select a supported input device and map armed audio tracks to input channels.
2. Pause the arrangement and position its cursor. For free capture choose Record armed tracks to disk.
3. Watch input meters. Capture does not automatically start arrangement playback or open a monitoring output.
4. Stop disk recording and wait for the writer to finish flushing.
5. Choose Recover and add to convert the captured data into owned audio assets and add the tracks in one arrangement edit.
6. Save the arrangement before dismissing a recovery entry.

Startup failure, disconnection, queue overflow, invalid samples or capacity limits can stop capture while retaining a recoverable prefix. Recovery requires the original project context, matching destinations and paused playback. Do not start another capture or attempt competing recovery while the existing worker is still tracked.

A crash can lose queued or unsynced audio. Recovery keeps complete frames and ignores torn final frames. Dismissing a recovery entry removes its raw spool/metadata, not already-created assets. Downstream browser waveform paths may still decode in memory; native disk capture does not make every editor unlimited.

1. Choose Timed punch or Timed loop takes, set the beat range and select 1–32 laps when looping.
2. Arm editable audio destinations with an empty range or one covering audio region. Pause conflicting Studio/arrangement playback.
3. Choose Render backing and record. Preparation makes a fixed backing with routing/automation; armed sources are silent in it.
4. Use direct monitoring and record after the ready lead-in. Do not change the prepared range during capture.
5. Recover the continuous recording and per-lap takes, audition alternatives, and select or comp the desired phrases.

Timed native capture currently requires the supported macOS duplex-device path with unambiguous input/output identity, a common sample rate and a compatible output buffer. It aligns using hardware callback timestamps, not an assumption that ADC/DAC and cable latency are zero.

Take selection retains alternatives. Overlapping comp fades are adjustable from 0–100 ms, with a 10 ms default in that control. Consolidation can prepare already-overlapping material for further comp edits while retaining originals. Listen for clicks and phase changes; an automated boundary test is not seam acceptance.

The timed range must fit the supported render limits; backing ends within one hour of arrangement zero, with an 8 MiB request limit. Long recovery may require temporary space for both continuous and separated lap assets.

Choose a native MIDI input and the intended instrument destination. Use count-in, timed capture, retained takes, retrospective MIDI, grooves and note tools where offered. Review recorded note timing and channels before committing a take.

Recorded expression includes member channels, modulation, sustain, breath, expression, timbre, 14-bit bend, channel pressure, per-note pressure and supported additional controllers. Editor lanes distinguish channel and per-note targets; trimming/playback chases state instead of assuming every clip begins with a new attack.

Saved MPE settings include zone, member count and manager/member bend ranges. Native instrument preparation sends ordered setup before notes. The plugin and physical controller must honor the same configuration; VST3 controller delivery depends on published mappings. This is MIDI 1.0/MPE support, not a blanket MIDI 2.0 translation promise.

Disconnects and overflow release retained notes/pedals. If expression sounds wrong, verify input channel/zone, pitch-bend range, instrument mapping and the selected lane before editing data to compensate.

1. Scan/select a supported installed plugin through the host controls. Native paths and entitlements are validated by the host.
2. Load an instrument source or insert, choose supported outputs and prepare playback.
3. For full preset editing, pause and choose Open editor. Edit the private draft instance, then Apply plugin changes to save one undoable state update.
4. Use Cancel to leave the saved state unchanged. Merely closing the vendor window can hide the draft; finish through Studio.
5. Save the project, reopen it and verify the plugin instance/state before relying on it in delivery.

The macOS editor paths include VST3 native views, AUv2 Cocoa views and AUv3 asynchronous view controllers. Missing/unsupported/timed-out views fail visibly; generic parameter controls remain useful. Native editor open/close is tier-gated. This does not certify every third-party plugin or operating system.

Record vendor automation during non-looping playback to capture supported changed normalized parameters through the playing identity-matched plugin. This is a bounded latch-style pass, approximately 20 Hz polling, not sample-accurate vendor gesture recording. Save/stop commits changed lanes; Discard restores underlying values. Opaque preset/topology changes belong in the paused editor.

In Plugin output buses and sidechains, choose the instance, main output and supported retained output routes. Additional outputs can route to destination buses. Validate the selected layout at native preparation; unavailable buses and feedback cycles are rejected rather than guessed.

An effect sidechain uses the chosen source after its inserts and fader/pan/mute, feeding the auxiliary input without summing into the target main input. The graph orders sources before dependents. A single-track freeze refuses external sidechains; use a session render or remove the dependency first.

AU/VST3 mono, stereo and supported 5.1/7.1 plugin layouts are adapted through the stereo mixer. Surround plugin output is folded down with the defined channel order and excludes LFE in that fold; this is separate from discrete surround delivery assignments.

Prepared graph paths support bounded delay compensation and host context, including tempo/meter, transport and supported loop/video timing. Dynamic latency/layout changes may invalidate preparation. Verify timing with the actual plugin; a plugin reporting a value is not proof of audible alignment.

Select a lane for gain, pan, mute, supported effect controls, sends or source timbre. Hosted paths bind plugin identity, slot and parameter; routing changes can invalidate a lane. Re-discover the intended target instead of transferring an old lane silently.

1. Start non-looping playback and choose Touch, Latch or Write.
2. Touch writes while held and returns to the original lane; Latch continues after first touch; Write begins immediately.
3. Finish the pass to save one undoable edit, or Discard to keep the underlying lane.
4. For multiple points, select them and move, offset values, duplicate, change curves or delete. Collision checks preserve neighboring points.

Linear, hold and eased baselines are supported with retained curve domains. Only one pass owns preview controls; transport/session discontinuities stop or discard it according to the operation. Recording is bounded to 36,000 stored points. Readback and vendor passes coalesce values and do not imply sample-accurate capture.

Source level/pan/detune and supported pulse-width/filter controls affect synthesized voices before inserts. They do not change recorded audio regions. Resetting optional timbre fields does not automatically remove existing automation lanes.

1. Open the asset library to search names/tags, mark favorites and find missing files.
2. Audition a bounded excerpt independently of arrangement playback. Apply BPM/root suggestions only after reviewing them.
3. Use relink review to match a replacement and apply it atomically to supported project references.
4. Export/import portable metadata when moving annotations. Keep the actual audio and provenance as well.

Tempo/key analysis is advisory. It examines a bounded excerpt and can abstain on silence, weak onsets or ambiguous tonal evidence. Half/double tempo choices and relative keys require judgment. Choosing a preview tonic does not rewrite project notes, tempo or scale.

Non-destructive sample edits create derived assets and retain restoration history. Reverse, normalize and supported range edits do not rewrite the original file. Long edits use bounded disk rendering; source, output and store limits still apply. Reopen and verify Restore before discarding originals.

MPC recipes preserve original sources/slices and selected edits. Portable kit import replaces the full pad set only after review; kit audio/recipes and session patterns/tempo have distinct persistence. Catalog/pack imports retain source credits, but declared rights remain the creator’s responsibility.

1. Select a supported track and stop conflicting operations.
2. Freeze to render its supported source/insert processing and automation into owned audio while retaining the original track snapshot.
3. Audition the frozen result. Unfreeze restores the saved source state when it remains available.
4. Choose Flatten only when you want an ordinary audio track without the retained unfreeze snapshot. Save an earlier project checkpoint first.

Flatten does not re-render or delete original asset files. Undo can restore its snapshot during the current history, but a reopened flattened project is ordinary audio and cannot Unfreeze. Freeze is bounded to the supported engine/plugin subset, saved state and source limits; unsupported sidechains or failures stop the operation.

Check tails, instrument expression, latency and bus routing with a reference mix. Supported long renders are disk-backed and capped; “streaming” does not mean arbitrary duration or unbounded storage.

1. In Compose + Deliver, stop playback, choose a loudness target and 0–30 second tail.
2. Choose Export master + bus stems and a destination folder.
3. Wait for render, file validation, stem-sum validation and loudness measurement.
4. Inspect delivery.json and listen to the master and premaster. Correct levels and export again if the target needs attention.

Stems sum to the premaster, not necessarily the nonlinear mastered output. Loudness presets in this export supply targets, not automatic normalization. A valid render outside target is saved with attention status; silence does not pass. The supported package is staged and published to a unique folder after checks succeed.

Current delivery limits include one hour, 64 native strips including master/extra outputs, 4 GiB aggregate PCM and bounded render/measurement deadlines. Keep free disk space and ensure the encoder/measurement executable is available.

Choose Stereo + 5.1 premaster or Stereo + 7.1 premaster and assign direct master contributors to front stereo, a speaker or mute. Unassigned contributions default to front stereo. Reconcile stale assignments after changing buses.

surround.wav uses 48 kHz / 32-bit float WAVEFORMATEXTENSIBLE and compensated premaster contributions. Stereo master effects/gain and stereo loudness targets do not certify this file. Single-speaker assignment averages L/R and can cancel out-of-phase material.

Live monitoring remains stereo. There is no automatic bass management, LFE filtering or surround loudness certification. Prepare an LFE bus explicitly and perform real speaker/listening acceptance.

1. Open Edit Video and choose local clips. Select a clip to trim source in/out, split, reorder or remove it.
2. Set cuts, fades or overlapping crossfades; choose canvas aspect and fit/crop behavior.
3. Import SRT for the selected source or add caption cues. Source-relative cues travel with clips; subtitle export uses assembled timeline time.
4. Choose one independent music track and set its trim, timeline offset and level. Set clip audio level to balance or replace original audio.
5. Save the project, collect its media if it must move, then export using a supported preset and inspect the finished file.

The editor is a single picture track with supported transitions, captions/titles and one independent music track, not an arbitrary multilayer compositor. Phone footage orientation uses validated display matrices. Unsupported mirrored/sheared or non-quarter-turn transforms fail visibly.

Music does not loop or extend the video. Preview and export apply fixed headroom based on clip/music levels; they do not automatically perform content-dependent loudness normalization. Browser preview clocks can pause while buffering; judge the exported media separately.

Picture hits can keep absolute session-second anchors while a suggested tempo span moves musical beats to them. Suggestions do not stretch audio or overwrite intermediate tempo changes. Drop-frame timecode affects displayed numbering and host context, not the underlying media frames.

Open Mixer control and recall, enable loopback OSC and/or choose a dedicated native MIDI input, then connect. Nothing connects automatically. Structural session changes, ownership conflicts, input overflow or a prolonged foreground polling gap require reconnecting.

OSC accepts bounded messages at the displayed loopback address with a per-connection capability. Keep it private. This is not a LAN listener, scheduled bundle engine or outbound controller feedback protocol. MIDI channels select the supported track slots; the panel maps gain/pan/mute/solo and program changes to recall.

Save up to sixteen named mix snapshots. Recall matches stable track IDs and preserves regions, plugins, routing and automation; authored automation retains priority. Mixer motion is not automatically recorded into automation lanes.

Open a factory pack or .gms source in the editor. Validate syntax and selected runtime capabilities before playback or export. MusicScript declarations do not serialize every recording take, plugin state, controller map or surround route.

MODULE loads a bounded .gmsmodule.json template and creates an editable source draft after confirmation. Transformations are declarative note operations, not arbitrary code or nested imports.

EXPORT GAME prepares supported loops, stems, transitions and a manifest. Middleware artifacts assist FMOD/Wwise import but do not automatically author every vendor transition, build SoundBanks or approve redistribution. Retain attribution and inspect loop seams in the target engine.

• MusicScript language and CLI manual
• libgms integration manual

Open from Web consumes a one-use expiring code for an owned cloud version. Review the compatibility report, import a copy and save locally. This is not two-way synchronization; keep the original browser version and a rendered reference.

For Go Meow Server, join the server session and arm one supported playback adapter. Every client needs local content; the server coordinates time, not audio/file distribution. Re-arm after edits or reconnects. MusicScript following requires the finite single-tempo supported subset and prepared local sample assets.

• Web platform manual
• Live-session setup

Finish acceptance with short controlled recordings, plugin state reopen tests, comp seams, MIDI/MPE expression, representative picture footage and exports on the intended hardware. Preserve each failing project and debug one issue at a time. Automated software tests do not replace these checks.

When reporting a problem, include build, OS/CPU, interface/rate, plugin version, public error and a minimal reproducible project with permission to share its assets. Never include license secrets or private access codes.