Media library

The media library is where every file enters Mosael: local uploads, screen/camera recordings, downloads from a link, and the output of renders, dubbing and AI generation. They all go through the same import path, so thumbnails, waveforms and duration probing are there no matter the source.

Documents such as PDFs, Word files, slide decks and spreadsheets come in here too and are parsed into readable text on your machine; see Documents.

Importing#

• Upload files (several at once), or drag them onto the window. Multiple files upload one after another; one failure does not stop the rest, and you are told how many came in and which did not, and why. QuickTime .mov videos such as Mac screen recordings are repackaged into an .mp4 container as they are imported — the picture and sound are copied as-is, not re-encoded, so quality is unchanged; scrubbing them in the app just stops stalling.
• Record the screen, camera, microphone, or screen and camera together; recordings land in the library.
• From a link — see below.
• From a web page — open a profile in the browser pool: the top bar's screenshot, video download and image collection save straight into the media library, as do files you download on the page, each remembering which page it came from and when.

Record screen and camera together#

Choose Screen + camera in Record media, select the camera and microphone, then start. Both captures start and stop together and have separate previews. Stopping creates two assets — a screen recording and a camera recording — rather than baking the camera into the screen video. They remain independently editable on the timeline. If either capture is empty, neither file is imported, preventing a partial pair.

Enable Mirror camera for a selfie-style composition. Both the camera preview and saved camera asset are flipped horizontally, while the screen asset stays unchanged. The choice is remembered for later recordings.

Mosael does not access the camera or microphone merely because the recorder was opened. After selecting a mode that needs those devices, choose Allow access to trigger the operating-system prompts. Once access is granted, selecting a mode with the camera opens it right away with a live preview (mirroring applies to the preview too), and recording takes over that same stream — so the dark first moments while a freshly opened camera settles its exposure are not recorded. Switching to a mode without the camera, or closing the recorder, releases it. Modes that record the microphone (camera, screen + camera, microphone) likewise open it once access is granted and show a live input level, so you can confirm there is sound before recording; the level stays in the floating controller while recording. A denied permission stays visible as an actionable state with links to the relevant system settings and a retry action. Screen and system-audio consent is requested by the operating system's sharing picker, as required by macOS.

Once capture starts, the recorder collapses into a controller in the bottom-right corner without covering or locking the app. You can browse media, enter the editor, or use other features while recording continues. When stopped, the files still return to the workspace or project where the recording was started.

System audio is enabled by default for screen recordings and captures sound playing on the computer. If the system sharing picker also presents a Share audio option, allow it there as well; turn the switch off for picture-only capture. If the picker returns no live audio track, Mosael stops the capture and asks you to retry instead of silently creating a mute asset. In Screen + camera mode, system audio stays with the screen asset while microphone audio stays with the camera asset, so both remain independently editable. Availability depends on the operating system and the capabilities exposed by its sharing picker. On macOS, a previously denied permission can be re-enabled from the recorder via Privacy & Security → Screen & System Audio Recording.

From a link#

Paste a video or playlist link, look at what is behind it, then decide what to download.

Probing before downloading is deliberate: a link may be one video, or a whole playlist — hundreds of items and tens of gigabytes. So pasting fetches metadata only (no media streams) and shows the listing: a single video is ticked for you, a playlist starts with nothing ticked.

• Audio or video is chosen before downloading. Someone who only wants the voice for transcription should not pay for hundreds of megabytes and a transcode.
• Max quality lists only the steps this link actually has — probing already knows what it tops out at, and offering a step that does nothing is letting the UI lie for the site. It is a ceiling, not an exact match: items in one batch don't all offer the same quality.
• Paging: 200 items at a time, with "Next" for the rest. A channel can hold thousands.

Content that needs a sign-in#

Members-only videos, private playlists and region-locked content read as "unavailable" without a session. This app already keeps every persistent login in the browser pool, so just borrow one of those identities — no need to export a cookie jar from somewhere else.

Available quality depends on the site, asset and sign-in state. Use the options returned by the current probe.

For multi-part videos and collections, every item in the listing carries its own title (the collection name is already in the header, so it is not repeated), and downloaded assets are named after it. When the site withholds a few names for a moment (Bilibili rate-limits bursts of requests), Mosael pauses and fills them in one at a time; any still missing read "Part 2 · name unavailable" or "Item N · name unavailable", can still be ticked and downloaded, and the asset is named after the real title the download brings back — never a whole list of "Untitled".

When nothing can be fetched, the message says which case it is: the site or link type is not supported, the link really has no video or audio, a sign-in identity is needed (pick a signed-in one under the identity selector above), the content is unavailable (private, removed or region-locked), or the network could not reach it.

Downloads run as a background job with progress in the task centre. One failure does not sink the batch: what downloaded stays in the library, and the failed count is reported.

Tags, filters and sorting#

• Filter by type (video / audio / image / document) or by tag from the toolbar. Several tags can be ticked at once; with two or more you choose between Has all of them and Has any of them. The library, the editor's media panel and the asset library share the same tag filter.
• Search looks at names, original file names and tags, ignoring case, as soon as you pause typing. Sort by Newest, Recently updated, Name or Duration; sorting by name interleaves Chinese names, by pinyin, with English ones and compares numbers by value (阿 < apple < Banana < 草莓, Episode 2 < Episode 10).
• Thousands of assets still open quickly. The list is fetched a page at a time, with filtering, search and sorting done on the server, and the next page comes in as you near the bottom; the grid and the list draw only the rows in view, and thumbnails load as they scroll in. The counts on the type tabs cover the whole scope (not the search). Select all fetches the remaining pages before selecting.
• Multi-select to tag or delete in bulk.
• Compare: multi-select assets of one kind to lay them side by side and pick a version. Images sit next to each other; videos share one screen and one play button, start muted (click a speaker to hear only that one), and frame stepping and timeline scrubbing apply to all of them at once.

Dub lines: a process's parts stay out of the way#

A subtitle dub makes one short audio file per subtitle, and lip-sync cuts a video into chunks to fix the lips — these are parts of a process: timelines and dub tracks still use them, but you don't need them when looking for something in the library, and there are often hundreds of them with the same name. So they are left out of the library by default, and out of the asset pickers, the editor's media pool and the agent's asset lists too.

Next to the type tabs on the media page, a Dub lines N chip (and Lip-sync chunks N when there are any) shows exactly those: a line at the top explains what they are, and Back to the library returns in one click. A dub line's card shows the sentence it speaks, and you can search by that sentence. It is only what you are looking at right now, not remembered; the count by the title is always the library's. When you upgrade, the parts that can be recognised (those remembering their dub sentence, synthesis started by a dubbing task, lip-sync chunks…) are marked automatically; anything that can't be told apart stays in the library.

Where an asset came from: provenance#

Every asset remembers where it came from. Cards and the detail view carry an origin tag: Imported, AI generated, Processed (made from another asset — trimmed, denoised, separated and the like) or Exported (a finished cut from the editor). The detail view's Made from row lists the source assets and the operation — trimmed, frame grab, grid split, to GIF, denoised, separated, export, joined, padded, mixed, from document, plugin — and you can follow it up level by level, then Back. A source that has since been deleted shows as "Deleted asset".

"Contains AI content" follows the chain: trim a piece out of an AI-generated video, or denoise an AI voiceover, and the new asset keeps the flag (an extra Contains AI content badge next to its tag) — even after the original is deleted. Export uses it to decide on the AI label; see Editing → Export.

Processing sound#

Audio and video assets have two extra items in their More and right-click menus. Both create new assets and leave the original untouched; they run as background tasks you can follow in the task panel. Right-clicking an asset in the editor's media pool offers the same two; to have the result swapped straight onto the timeline, right-click a clip there and choose Denoise (replaces when done), Keep only the voice or Split into voice and background (see Editing).

• Reduce noise…: pick a method and a strength. A cleaned video keeps its picture; only the sound changes. The methods trade off differently:

  Method	Good for	Music	Setup
  Built-in	Steady noise: air conditioning, fans, hum	Untouched	None
  DeepFilterNet	Speech-led recordings, including keyboards, dishes and chatter; the most natural voice	Removed	A deployment admin downloads it once under Admin → Engines (about 30 MB)
  RNNoise	Same, lighter and not as good as DeepFilterNet	Lowered	None

  Built-in is the default, because "reduce noise" rarely means "and drop the music". It measures how loud the noise is before it acts, so one strength suits faint and heavy hiss alike; for speech-led material DeepFilterNet does noticeably better. Keeping only the voice and nothing else is not noise reduction: use Separate voice and background below and take the voice stem.

• Separate voice and background: split into a voice asset and a background asset. The background is everything except the voice: music, ambience, effects. The engine runs on this machine and must be installed first by a deployment admin under Admin → Engines (only deployment admins install engines; other members see “installed by a deployment admin” where it is needed); a long asset takes many minutes on a CPU. That section reports “Installed” only after actually loading the engine once — not by looking for a directory; if an install was interrupted it says the runtime is incomplete, and clicking Install again adds the missing dependencies.

Workflows have matching Reduce noise and Separate voice and background nodes (under Audio), and the agent can do both too — it opens a confirmation card first.

Convert video to GIF#

Right-click a video and choose Convert to GIF. Frame rate, width, start time and clip duration are configurable. Conversion runs as a background job and registers a new GIF with lineage back to its source; the original video is never changed or overwritten. The same operation is available as a Video to GIF workflow node.

Preview and details#

Click an asset to open its preview. Video has dedicated playback controls; audio shows a waveform and progress; images open directly. Names, types and file details sit beside the preview. Long titles and IDs wrap within the panel.

The filter toolbar keeps search and filter controls aligned. Its translucent background changes smoothly while scrolling. Only one card menu opens at a time; click outside or press Escape to dismiss it.