Skip to main content
All releases

v2.4

June 24, 2026

Per-Stream Capture Quality, Screen Bitrate Control, and Custom Output Folders

What's New

Per-Stream Capture Quality (Screen and Camera)

A screen and a moving camera image compress very differently, so a single bitrate rarely fits both. In Advanced capture quality, you can now configure the screen and camera streams independently.

  • A new Screen / Camera switch sits at the top of the Advanced controls — in both the source sidebar and the Output panel. Pick a stream, then set its codec, bitrate, profile, and keyframe behavior on its own.
  • The High, Max, and Compat quality profiles still apply to every stream at once; only Advanced lets the two diverge, so simple setups stay simple.
  • Camera resolution and frame rate stay controlled by each camera's own settings, so the per-stream panel focuses on codec and bitrate.
  • Each stream's profile is remembered separately. The camera profile starts from your existing settings, so nothing changes until you customize it.

Record Into a Custom Folder

Send a recording's files straight into a folder you choose — for example the project folder where you keep that video's scripts and code — instead of only the single global output folder.

  • A Record into a specific folder control sits in the Session panel, right next to the project and output path.
  • The choice persists until you clear it, so it survives relaunch and multiple takes of the same video land together. Use Default returns to the standard output folder.
  • Sessions recorded into a custom folder stay first-class: they list, export, transcribe, reveal, rename, clean, delete, and crash-recover like any other recording, even after relaunch.
  • The sessions list is grouped by project, not by folder, so all of a project's takes stay visible together — and every row shows a light hint of which folder it was recorded into.

One-Tap Preview Toggle

A single Preview On / Preview Off control now sits directly above the previews, so you can stop and start the live preview without opening Settings.

  • Turning the preview off stops Lumary's preview rendering, lowering CPU and GPU use while you set up sources you don't need to watch live.
  • It reuses the existing idle-preview pipeline, so your recording sources and output settings are untouched.

Fixed

Video bitrate now applies to screen recordings

The Advanced video bitrate only changed the camera stream; screen recordings ignored it and were captured near-lossless, producing very large files — about a gigabyte per minute for a high-resolution display.

  • An explicit bitrate is now honored as a real ceiling for screen capture, not just camera. The near-lossless quality override is no longer applied once you set a bitrate.
  • Default quality profiles are unchanged: if you have not set a custom bitrate, screen capture still uses Lumary's resolution-aware default for crisp text.

Paired camera missing from Resolve and Final Cut timelines

When a camera is paired with a microphone, Lumary records it into a single combined file, so the camera lived in the session's paired outputs rather than its standalone source list — and the FCPXML timeline export dropped it.

  • The timeline export now places paired camera and display outputs, each on its own track.
  • To avoid doubled audio, a paired file contributes only the media not already on the timeline as a standalone source — video-only when the paired microphone is also recorded on its own, video and audio when it is not.
  • Single-session, per-session, and concatenated multi-session timelines are all covered, and the export still references captured media directly — no transcode, no muxing, no sidecar files.

Transcription recovers from an incomplete model download

Transcription could fail with a low-level "tokenizer.json missing" error, and Try Again hit the same error every time.

  • Lumary now validates a model's Core ML components before treating it as downloaded, so an incomplete model shows as not downloaded and prompts a clean re-download.
  • The speech tokenizer is fetched at download time and cached alongside the models, so transcription no longer needs a network round-trip later.
  • Re-downloading clears any partial files first, and missing files now surface a clear, actionable message instead of a low-level error.

Requirements

  • macOS 15.0 (Sequoia) or later
  • Stream Deck 6.9+ (for Stream Deck integration)