The main window in its empty state, shown before any media has been added. It explains what the application produces — subjects, on-screen text, speech and shot framing turned into Final Cut Pro keywords — and provides the entry point for loading media. The same window becomes the asset list once files are added; the action bar remains available at all times.

• Media drop area — the central dashed region, labelled "Drop video, audio, images or folders — or click to browse". It accepts files and folders dragged from Finder or another application. Dropping one or more items (or clicking the region to open a file chooser) adds the media to the working list and begins reading their metadata. Video, audio and still-image files are accepted; a dropped folder is scanned recursively and every supported file inside it is added, with the folder hierarchy preserved and later used to derive folder-name keywords.
• Supported format caption — lists the file types the drop area recognises: MOV, MP4, M4V, MPEG, WAV, AIFF, MP3, M4A, AAC, JPG, PNG, HEIC and TIFF. Files outside this set are ignored during a folder scan.
• Cumulative Tags — combines the keywords of every file currently in the list into a single aggregate tag set and opens it for inspection, giving a shoot-wide overview instead of a per-file one. Unavailable while the list is empty.
• Analyze — runs the enabled analyses across every file in the list: subject recognition, on-screen text recognition, speech transcription and shot-framing classification, plus creation-date and folder-name tagging. Progress is reported per file and overall. It stays unavailable until all added files have finished loading, and remains disabled for the whole run so a batch cannot be started twice.
• FCP 10.6+ — indicates the Final Cut Pro version the export currently targets, which determines the FCPXML schema variant that will be written. The value follows the configured export version.
• Export FCPXML — produces the tag data for the current list and delivers it either as a saved .fcpxml file or directly into a running Final Cut Pro session, according to the configured export mode. The label reflects the chosen mode. It is unavailable while the list is empty and while any file is still awaiting analysis, since only completed files can be exported.

The main window manages one working batch of media assets. Files are ingested here, each file's four analysis engines are enabled or disabled, the batch is analysed, and the resulting keywords are synthesised across the whole batch or exported to Final Cut Pro. The window is split into a scrollable asset list (with a pinned header) and a permanently visible action bar at the bottom.

Drag area:

• The entire window accepts dragged files and folders. Dropped items are added to the batch; folders are scanned recursively and hidden items ignored.
• Accepted media: video (mov, qt, mp4, mpeg, m4v), audio (aif, aiff, wav, wave, mp3, aac, m4a), stills (jpg, jpeg, tif, tiff, png, heic).
• While a drag hovers over the list, a dashed accent overlay reading "ADD MEDIA / Release to add to the list" appears; releasing the drag appends the resolved assets to the batch.

1. Batch Summary and Add Media

• Batch summary — "6 Assets · 0:08:39" states how many files are loaded and their combined running time, giving an immediate sense of the workload before analysis.
• Add Media (+) — opens the system media picker for files and folders, adding everything selected to the same batch. Equivalent to a drop.

2. Column Analysis Toggles

Four icons aligned with the four analysis columns below them. Each one switches that analysis on or off for every file in the batch that supports it; unsupported files are left untouched.

• Subjects — visual subject recognition on video and images.
• On-Screen Text — optical character recognition of visible text on video and images.
• Speech — transcription and keyword extraction from the audio track of anything carrying one.
• Shot Framing — classification of the dominant camera framing, video only.

Each shows an aggregate state: filled accent when on for every supported file, hollow when off everywhere, partially filled when the batch is mixed, and a plain dot when no file supports it. Turning a column back off clears that analysis for the whole batch, letting a run be narrowed without editing files individually.

3. Per-File Analysis Toggles and Status

Each asset row repeats the same four analyses as independent switches for that single file, so a batch can mix engines.

• Poster / type tile — thumbnail of the frame, a waveform tile for audio files, or a type glyph.
• Title and detail line — file name plus duration, frame rate, source sub-folder and tag count once available.
• Per-file analysis switches — grey hollow = off; accent-filled = enabled and queued; a filling accent ring = currently running; a green check = finished; a dot = not applicable to this media type.
• Status indicator — Loading while metadata is read, Ready when idle, Queued when waiting its turn in a run, a percentage while analysing (shown here at 64 %), and Done with a checkmark once every enabled analysis for that file has completed. An inline progress bar under the file name tracks the same run.
• Right-clicking a row offers viewing its tags, revealing it in Finder, enabling or disabling every analysis for that file, and removing it from the batch; double-clicking opens its full tag set.

4. Action Bar

• Cumulative Tags — merges the keywords of every file in the batch into one synthetic set (subjects, text, speech, framing, dates, folders, generated and personal tags, with transcriptions re-processed and deduplicated) and opens it for inspection. Useful for judging a whole shoot at a glance.
• Batch progress with percentage — the aggregate advancement of the running batch (27 % here); it appears only while analysis is in flight.
• Analyze — the primary action: starts the enabled analyses across the whole batch. It is unavailable while files are still loading or a previous run is in progress, and is highlighted when files are waiting, marking it as the obvious next step.
• FCP version label ("FCP 10.6+") — the Final Cut Pro generation the export will target, which decides the XML schema produced.
• Export FCPXML — writes the tagged assets and their keyword collections as Final Cut Pro XML, or sends them straight into a running Final Cut Pro session, depending on the configured delivery mode. Unavailable until the batch is empty-free and fully analysed.

Presents the complete analysis outcome for a single media asset. The window combines a playback/inspection preview, the file's technical identity and processing status, and the full set of keywords the analysis produced, grouped by source category. From here the user reviews, prunes, extends and copies the tags that will be written to Final Cut Pro and to Finder.

The interface is divided into the following numbered topics:

1. Media Preview

The visual or audible representation of the asset being reviewed, spanning the full width of the upper area.

• Video assets play inline with floating playback controls.
• Audio assets are represented by their waveform, drawn in the application accent colour; clicking anywhere on the waveform jumps playback to that position.
• Still images are shown aspect-fitted on black.
• The aggregate "cumulative" tag set (all files combined) is shown as a generic stacked-tile graphic, since it has no single source file.

2. File Identity and Analysis Status

The header directly beneath the preview establishes which asset the tags belong to and whether its processing has finished.

• The file name identifies the asset being inspected; the entry is truncated in the middle so the extension remains readable.
• The detail line reports the asset duration as a timecode and its frame rate, giving editorial context for the tags below.
• The status pill on the trailing edge reflects the analysis lifecycle: still loading, ready, queued, analysing with a running percentage, or done. "Done" confirms every enabled analysis has completed and the tag set is final.

3. Tag Summary and Layout Presentation

The band separating the header from the tag body summarises volume and controls how the tags are displayed.

• The total tag count states how many keywords currently apply to this asset across all sources.
• The Cloud / Categories selector switches between a single undifferentiated flow of all tags and a grouped presentation with one block per source category. Only the selected presentation is persisted; Cloud is the default.
• In grouped presentation, each block carries its own icon, name and count, and the personal tag block is always present even when empty, so there is always a fixed place to add a new tag.

4. Category Filters

A row of category pills, one per source that produced at least one keyword, showing the count contributed by each.

• Each pill is colour-coded to match the tag colour used for that source, linking the filter visually to the tags it governs.
• Toggling a pill off hides that category's tags from the body; toggling it back restores them. Filtering is per window and never alters the underlying data.
• The visible/muted state of each pill makes it obvious which sources are currently contributing to the displayed set.
• Subjects and Speech derive from visual subject recognition and speech transcription respectively; Shots reflects camera framing classification; Date reflects the creation timestamp formatted according to the configured date pattern.

5. Tag Set

The main scrollable area holding every keyword for the asset, each rendered as a colour-tagged entry.

• A leading colour dot identifies the originating category when the Cloud presentation is used; in Categories presentation the enclosing block supplies that context instead.
• Pointing at a tag reveals its removal affordance; the space for that affordance is reserved at all times so the flow of tags never shifts under the pointer.
• Removing a tag deletes every occurrence of that keyword from the corresponding source list on the asset, immediately updating the tag count and any category filter counts.
• The context menu on a tag offers copying the single keyword to the clipboard and removing it.
• The trailing dashed "+ Add Tag" entry converts into inline text entry: Return commits the keyword to the personal tag set, Escape discards it, and leaving the field also commits any non-empty text.
• Duplicate or near-duplicate entries are visually de-duplicated by category and keyword so the same tag does not appear twice.
• An empty body states why nothing is shown — analysis in progress, analysis complete with no keywords found, or analysis not yet started — and always offers the add-tag entry so the user can still annotate manually.

6. Actions

The trailing bar exposes the two inspection views and the primary output action.

• Audio Tags opens the audio keyword manager, where spoken words are triaged by frequency, excluded from future results, or promoted to triggers. It is unavailable until speech results exist and the asset's analysis is complete.
• Full Transcription opens the complete dialogue transcript for review and copying. It is unavailable while no transcript exists.
• Copy Tags places the entire final keyword set on the system clipboard as a comma-separated list and confirms how many tags were copied. It is unavailable when the asset has no tags.

Semantic impact summary: every tag shown here originates from a distinct detection engine — visual subjects, on-screen text, speech, shot framing, creation date, folder path, generated rules, or manual entry — and each is stored on the asset separately. Edits made in this window therefore change what is exported to Final Cut Pro as keywords and what is written back to Finder tags, without re-running any analysis.

Presents the analysis outcome and curatable keyword set for a single media asset, together with its preview and the follow-up actions available for the extracted metadata.

• Media preview — for an audio asset, the full waveform is drawn over a tinted backdrop. Clicking anywhere on the waveform moves playback to that time; a thin vertical marker follows the current playback position. The inline transport strip beneath provides play/pause, elapsed and remaining time, position seeking, and playback-output selection.
• Asset identification — the file name, a waveform symbol with the total duration as a timecode, and the current processing state (green "Done" once every enabled analysis has finished). While an analysis is in progress this area reflects progress instead.
• Tag total — the number of keywords currently collected for the asset.
• Category filters — one capsule per non-empty tag category, each showing its symbol, name and keyword count (for example Speech 5, Date 1). Selecting one hides or shows that whole category in the listing below; hidden categories are drawn faded. Each also offers Copy and Remove through its context menu.
• Keyword listing, grouped by category — each block shows an icon, an uppercase category name and its count:
• Speech — keywords derived from the spoken content, each presented as a removable capsule (for example cattolicesimo, peccato, mangiare, bevono, deboli).
• Creation Date — the asset's creation date formatted according to the configured date pattern (for example 23 9 2026).
• Your Tags — manually authored keywords; always shown, even when empty, so there is a stable place to add one.
• Removing a keyword — pointing at a keyword reveals a removal affordance; using it deletes every occurrence of that keyword from the asset. A right-click offers copying the individual keyword or removing it.
• Add Tag — turns into a text entry field; confirming appends the typed value to Your Tags (duplicates and blank entries are ignored), while dismissing cancels the entry.
• Audio Tags — opens the keyword manager for the spoken content, where occurrence thresholds and ignore/trigger membership can be adjusted. Unavailable until the speech analysis has produced results.
• Full Transcription — opens the complete raw transcript of the audio in a separate viewer. Unavailable while no transcript exists.
• Copy Tags — places the complete keyword set for this asset on the clipboard as a comma-separated list and confirms how many tags were copied. Unavailable when the asset has no keywords.

Controls how media analysis produces keywords and how those keywords reach Final Cut Pro and the Finder. Each option applies to every subsequent analysis and export.

Tag Sources

• Include Folder and Subfolder Names — when on, the names of the folders containing an asset, relative to the dropped root folder, are added as keywords.
• Include Creation Date — when on, the asset's creation date is converted into a keyword.
• Format — the date pattern applied to that creation date (e.g. day, abbreviated month, year). Ignored while the creation date keyword is off.

Tag Manager

• Ignore and Trigger — when on, the ignore and trigger rules are applied, so listed words are either dropped from results or forced to appear regardless of occurrence count.
• Generator — when on, the Boolean generation rules run and can add derived keywords to assets.
• Tag Manager — opens the rule editor where the ignore, trigger and generator lists themselves are defined.

Finder Tags

• Save Video Tags — when on, visual and framing keywords are written to the file's macOS Finder tags after analysis.
• Save Text Tags — when on, on-screen text keywords are written to the file's macOS Finder tags.

Export

• Save FCPXML File — export produces a standalone .fcpxml document for manual import or archiving.
• Send Directly to Final Cut Pro — export injects the assets and keyword collections into a running Final Cut Pro session.
• Final Cut Pro 10.6 and Following — when on, the export targets the modern schema for Final Cut Pro 10.6 and later; when off, it targets older versions up to 10.5.x.

Controls the frame sampling density and the confidence/count limits used when Vision subject recognition runs on video and still assets. Configuration here determines how thoroughly each file is scanned and how many subject keywords reach the exported tag set.

Frame Scanning

• Scan — sets how densely frames are sampled for subject detection, expressed as one analysed frame every N frames (nine granularities, from every frame up to 1 in 300). Higher values analyse fewer frames, reducing processing time on long clips at the cost of missing brief subjects; the current readout shows the selected step and the caption below states the equivalent ratio in plain language ("1 in 300 Frames").

Subject Filtering

• Threshold — the minimum confidence percentage a detection must reach to be kept as a subject tag (1 %–30 %). Lower values yield more tags and more false positives; higher values yield fewer, more reliable tags and a cleaner keyword set. The stored value is normalised internally, so the displayed percentage is the human-facing form.
• First — how many of the highest-confidence subjects are always retained regardless of other limits. Guarantees that the most salient subjects are not crowded out when the maximum count is restrictive.
• Limit Subject Count — enables or disables a hard cap on the number of subject tags produced per asset, preventing keyword explosion on visually busy footage.
• Max — the maximum number of subject tags retained per asset. Only takes effect while the count limit is enabled; with the limit off, every qualifying subject is kept.

Configures the speech-to-text pipeline that converts spoken audio into keywords for Final Cut Pro. The values here govern the accuracy, volume and reach of the tags produced from every analysed video or audio asset.

• Language — Sets the locale (currently Italian (Italy)) whose recognition model is used when transcribing speech. Changing it affects all subsequent analyses, so a mismatch produces poor or empty keyword results.
• Min. Occurrences — Sets how many times a word must be spoken before it becomes a keyword (currently 4). Raising the value keeps only recurring topics and removes one-off filler; lowering it surfaces more words at the cost of noise. Words that are designated as triggers bypass this threshold.
• Include Triggers — When on, any word registered in the trigger list is always emitted as a keyword, even if it appears fewer times than the threshold allows. Useful for guaranteeing rare but essential terms such as product names, client names or camera cues.
• Enable Shared Transcription — Off by default. When turned on, transcription work is distributed over the UMAN network: segments may be processed on other machines and this Mac in turn contributes processing power. Activation requires explicit confirmation, and it only applies when the built-in on-device engines are not used.