Works & Tasks

Assets, Characters, Voice Library, Music Library, Plugins, and the Job Center: outputs, sources, and files all stay local.

The Resources group in the sidebar has four pages: Assets, Characters, Voice Library, and Music Library. Plugins live under Settings → Plugins. Jobs live under the Jobs item in the sidebar.

Assets

Successful outputs. You can switch pages while generation runs; the job keeps going. The same browser tab keeps showing progress; after you close the tab, the progress bar is no longer attached, but the job still runs.

Where assets come from:

• Generation results from each function page go straight into the Assets library.
• Image/video/audio concat, Image Enhance, and Video Enhance on the Tools pages put their outputs straight into Assets; no extra save step.
• Function, speech, and concat nodes on the canvas are not listed by default; click "Save to library" in the top-right corner of the node preview to make them appear here. Turn on "Include canvas intermediates" to see these unpublished files and masks.
• Files uploaded directly through input nodes, plus new images exported from markup or crop, become normal assets.

Depth map results carry a depth map marker, and the filename ends with _depth.png; for example, photo.jpg becomes photo_depth.png.

Characters

Sidebar Resources → Characters. Stores a character's name, appearance description, and one 16:9 four-panel character sheet for image-to-video, Digital Human, and Motion Transfer. See Video → Characters for generation from a description, uploading a character sheet, automatic splitting, and what each panel is used for.

Voice Library

Sidebar Resources → Voice Library. Stores the reference audio and reference text for cloned voices; Text to Speech and Voice Conversion pick voices here. See Speech → Voice Library.

Music Library

Sidebar Resources → Music Library. Upload, preview, and delete BGM; duplicate names get an automatic suffix. Quick Create and Custom Assets pick background music here. See Speech → Music Library.

Plugins

Settings → Plugins manages only local add-on services: Liber Speech (port 5555) and Liber Depth (port 5556). You can search. Cards show an icon, name, and description; open a card to see paths, start/stop, and installation.

When connecting to a speech or depth service that already runs on the LAN, do not install it here. Enter its address under Service Providers.

Installation. After you click "Install", the toggle becomes a blinking "Installing", and both the card and the details show a progress bar. The stages are: prepare the Python environment, install PyTorch, install the plugin's own dependencies; the depth plugin also downloads its default model. You cannot enable it before installation finishes. On failure, the details show the error and the progress bar disappears.

Speech and depth share one Python environment with PyTorch. When that environment is already detected, a later plugin only adds the packages it is missing and does not install PyTorch again.

Startup	Path
Start from backend in development	Environment at %LOCALAPPDATA%\LiberAstra\plugin-runtime. Models at backend/data/plugins/<speech or depth>/models
Windows installer	Data directory is %LOCALAPPDATA%\LiberAstra\data. Environment at plugins\runtime inside it. Models at plugins\<speech or depth>\models

If a released LiberSpeech installer includes docker-compose.yml and install.ps1, enabling it uses Docker and does not create the Python environment above. services/speech in this repository has no install.ps1, so it still uses Python.

Enable and uninstall. Enabling starts the local process and adds a matching entry under Service Providers (Liber Speech API or Liber Depth). Disabling stops the process and removes that local entry. Uninstalling also stops the process. Docker-based speech runs docker compose stop when disabled, not down.

Speech weights are not downloaded during installation; they land in the speech model directory on first synthesis. Depth installation downloads only dpt_swin2_large_384.pt; see Image → Depth Map Generation for the other weights.

Jobs

Sidebar Jobs. Shows progress and status. "Retry" reuses the old parameters; after changing settings, submit a new job. Function nodes, speech nodes, and concat nodes on the canvas, plus concat (media.concat) and Video Enhance (video.v2v) jobs from the Tools pages, also appear here.

Switching to another page while a job runs does not interrupt it; return to the Job Center to keep watching progress. To stop it, cancel the job. The canvas "Run" button turns into a red "Stop" while running; click it and confirm to cancel the current orchestration, stop downstream nodes, and send an interrupt request to ComfyUI.

---

Related: Settings → Workflow Packs; Settings → System Settings.