Shadow Creator Studio

A Linux desktop app for recording professional YouTube videos — screen, webcam, tutorials, and voice — with a simple creator-focused interface. The long-term engine is OBS Studio (via WebSocket) with FFmpeg + NVIDIA NVENC as the tooling and fallback path.

Screen and Presentation record through the xdg-desktop-portal ScreenCast session plus GStreamer pipewiresrc. Camera / Voice / Creator still use FFmpeg. OBS WebSocket is an optional engine when OBS is already running. Cloud AI and YouTube OAuth stay disabled.

Screenshots

Add PNG captures under docs/screenshots/ after you build the GTK shell (libgtk-4-dev and libadwaita-1-dev).

Screen	Placeholder (not captured yet)
Record	docs/screenshots/record.png
Settings	docs/screenshots/settings.png
Diagnostics	docs/screenshots/diagnostics.png
First-run wizard	docs/screenshots/wizard.png

What works

• Everything from Milestone 1 (shell, settings, wizard, diagnostics, dashboard)
• Camera selector with name, resolution, and pixel format (not raw /dev/videoN as the only label)
• PipeWire microphone and desktop-monitor lists from pw-dump
• Live camera preview in the large preview area, with an optional mirror flip
• Real-time mic and desktop meters (peak + average) plus a clipping warning
• Multi-monitor display tiles from the session; window capture uses the portal picker; region stays labeled unavailable
• Last camera, mic, desktop source, display, and recording mode persist in settings
• Background discovery — the UI does not freeze while probing devices
• START RECORDING for Camera, Voice, Creator, Screen (portal desktop), and Presentation (desktop + webcam PIP)
• Separate tracks: mixed + mic + desktop + optional music, each with mute and volume
• Mic chain in FFmpeg: HPF → afftdn → gate → EQ → compressor → limiter
• Presets: Natural (default, light), Podcast, Broadcast, Quiet Room, Noisy Room, Voice, Raw
• Calibration from the live peak meter (recommend gain + a light preset)
• Mic monitoring toggle does not create a speaker loopback (feedback-safe)
• Library page indexes the configured recordings folder (not the whole disk)
• Play, open folder, rename, remux, export, copy path, delete with confirm
• Quick edit writes a new file: trim, normalize, 720p, compress, extract audio, MP3, WAV, GIF, thumbnail, YouTube-ready MP4
• Teleprompter: paste script, font size, scroll speed, pause, mirrored, overlay window
• In-app hotkeys (F9 start/stop, F8 marker, F7 mute, F6 camera, F5 teleprompter)
• Markers write *.markers.json beside the take for later chapters
• Local extensions: thumbnail, chapters-from-markers, 9:16, silencedetect, silence-remove
• Whisper only if a local binary and an already-downloaded model exist
• Disabled tiles: captions, highlights, Shorts, AI title, YouTube upload
• Crash-safe MKV names like 2026-09-20_YouTube_Record_001.mkv — never overwrites
• Hardware encode when FFmpeg lists NVENC (H.264 / HEVC / AV1); otherwise libx264
• Quality presets: YouTube Standard / High Quality / 4K, Archival, Small File, Custom
• Optional copy remux to MP4 after stop. The MKV is never deleted
• Live timer, real encoder name, disk warnings, stop confirmation

What is intentionally unavailable

Control	Label
Region capture	Not offered by the COSMIC ScreenCast portal
Global compositor hotkeys	COSMIC has no GlobalShortcuts portal. Use in-app keys or shadow-creator-studio --action start-stop
Mic speaker loopback	Never created by default. Headphones + desktop mixer, or an explicit later opt-in
YouTube OAuth / cloud AI	No hardcoded credentials
Whisper without a local model	Nothing is downloaded automatically
GO LIVE	Needs secret-tool + a stored key + an flv/RTMP path. The key is never logged

Dependencies

Typical Pop!_OS 24.04 creator desktop

• Rust 1.80+ (rustc, cargo)
• GTK4 4.14+ and libadwaita 1.5+ runtimes to run the window
• libgtk-4-dev and libadwaita-1-dev to compile scs-ui
• PipeWire (Pulse compatibility is fine)
• FFmpeg 6.x; NVIDIA NVENC encoders when an NVIDIA GPU is present
• gstreamer1.0-tools plus pipewiresrc plugins for Screen / Presentation
• secret-tool (sudo apt install libsecret-tools) to store stream keys
• Optional: OBS Studio with obs-websocket enabled; EasyEffects

Foundation crates and tests compile without GTK headers.

Install

git clone https://github.com/Shadowfetchapps/shadow-creator-studio.git
cd shadow-creator-studio
sudo apt install build-essential pkg-config libgtk-4-dev libadwaita-1-dev \
  gstreamer1.0-tools gstreamer1.0-plugins-good gstreamer1.0-pipewire \
  ffmpeg pipewire-bin libsecret-tools
cargo build -p scs-ui --release

User-local desktop entry (no sudo): see packaging/README.md. Optional .deb recipe: ./packaging/make-deb.sh after the release build.

This project never runs sudo for you. Full package list: BUILDING.md.

Run

Libraries and tests (works without GTK headers):

cargo test --workspace --exclude scs-ui --lib

(cargo test also works; system rustdoc on some Pop!_OS images is missing LLVM and will fail doctests only. --lib skips that.)

GUI (needs GTK4/libadwaita dev packages and a display):

cargo run -p scs-ui

Config is written to:

$XDG_CONFIG_HOME/com.shadowfetch.creatorstudio/settings.json

Default recording folder (created when you choose it / finish the wizard):

$XDG_VIDEOS_DIR/Shadow Creator Studio

Recording notes

• Target container: MKV (crash-safe). Optional remux to MP4 without re-encode starts in M6. The MKV is never deleted until the MP4 verifies.
• Creator muxes mixed / mic / desktop / optional music as configured. Filter graphs use pad labels only.
• Preferred future encoder: NVIDIA NVENC (h264_nvenc, plus HEVC/AV1 when FFmpeg lists them).
• Primary future record engine: OBS via obs-websocket. FFmpeg is remux / probe / fallback. OBS is not required to launch the studio.

Audio

• PipeWire is the native path. pactl showing “PulseAudio (on PipeWire)” is normal.
• Mic chain (M4): high-pass → afftdn denoise → gate → EQ → compressor → limiter. Default Natural (light). Not written into EasyEffects or WirePlumber.
• EasyEffects is compatible in concept and optional. Shadow Creator Studio will not silently change system mic routing.

NVIDIA

On a machine with an NVIDIA driver and NVENC-capable FFmpeg, the dashboard reads GPU load, VRAM, and temperature through NVML when initialization succeeds. It does not spawn nvidia-smi every tick. While recording, the status chip shows the FFmpeg encoder actually in use.

Known limitations

• Region capture is unavailable on the COSMIC portal.
• FFmpeg has no PipeWire demuxer here; desktop capture is portal + GStreamer, not x11grab.
• OBS WebSocket records the current OBS scene. This app never writes OBS configs or secrets into the repo.
• GO LIVE needs secret-tool. If it is missing, install libsecret-tools.
• Global hotkeys are in-app, plus --action for COSMIC Custom Shortcuts.
• Camera preview pauses while an FFmpeg camera take owns V4L2.
• The app never writes WirePlumber, EasyEffects, or default source/sink configuration.

License

MIT.