Cutwise had reached the point where adding another feature also meant finding somewhere sensible to put its controls. The timeline could already hold cuts, captions, images, callouts and extra audio, while Shorts had introduced a second place to frame and export video.
The next round of work, on 6 October, was about making those pieces easier to use together. I reorganised the editor, added corrections to the transcript, gave Shorts more direct framing controls and built a way to cover parts of a recording before exporting it.
In the previous post, I described the first AI Shorts workflow and the Windows investigation. These changes build on the editing side of that work. The Windows failures still have their own outstanding fixes and validation.
Giving the edit more room
I reorganised the main workspace around four stages: Prepare, Edit, Review and Export. The application opens in Edit, with the video, transcript and timeline together. Preparation, suggestion review and export settings now have separate panels.
On a wider window, the preview stays visible while the timeline scrolls underneath it. A draggable divider changes how much space each receives. I also made the timeline controls more compact and moved detailed settings into inspectors, so the growing collection of options takes less room away from the recording.
Prepare can run local transcription when it is needed, followed by local cleanup. Optional cloud analysis still requires its transcript disclosure, and the resulting suggestions wait for review. I added editing presets so cleanup settings and instructions can be reused on another recording.
The review panel gained a keyboard mode: A accepts, R rejects, N and P move between suggestions, and Space builds a comparison preview. In the editor, Space controls preview playback. Those shortcuts respect text input, so correcting a sentence does not accidentally start the video or accept an edit.
Correcting the words without moving the recording
An inaccurate transcript affects more than the text beside the player. It can feed the wrong wording into captions and later AI analysis, even when its timestamps are useful.
I added an Edit text mode for correcting individual transcript words. The correction is stored separately from the original transcript snapshot and leaves the word's timing intact. New analyses and generated captions use the corrected text, while ordinary click-to-seek navigation returns when editing is finished.
Existing captions need more care. A caption block may already contain a deliberate rewrite, different timing or a line break chosen during review. Automatically regenerating it after every transcript correction would throw that work away.
Instead, Update captions from transcript is an explicit action. It explains that the main video's caption text and timing will be replaced, keeps the caption style and supports undo. Saved Shorts retain their own transcript snapshots.
That gives the transcript and captions a useful relationship without treating them as interchangeable copies of the same text.
Making a Short from a moment I already know
AI suggestions are useful when I want help finding an excerpt. They are an unnecessary step when I already know which part of the recording I want.
I added Create short from selection. A source range selected on the main timeline can go straight into Shorts, and a Shift-click selection of transcript words provides another way to choose the passage. The timeline route works without a transcript or cloud credentials.
There are still boundaries to respect. A selection whose start or end falls inside removed footage is rejected, and repeated source footage uses its first retained occurrence. Once created, the Short opens for framing and export with its own saved editing snapshot, just like an AI suggestion.
I also tightened up saving. Valid changes save automatically after a short pause, and preview, export, project switching and normal exit wait for the save to succeed. Invalid values remain visible for correction or can be reverted. If saving fails, the editing buffer is retained for another attempt.
That matters most when several small changes happen together. Adjusting the title, timing and crop should not leave the export using an older version of one of them.
Letting the portrait crop move
The first Shorts version had Fit and Fill modes, with position and zoom controls. I expanded that into an interactive framing canvas: drag the picture inside the portrait frame, scrub to another moment and reset the crop when needed.
A single crop still cannot suit every moment of a screen recording. The relevant part of a demonstration might move from one side of the screen to the other, or need a closer view for a few seconds.
I added crop keyframes in Fill mode. A keyframe records the position and zoom at a chosen moment; preview and export interpolate between them, holding the first and last framing outside their range. Once keyframing is enabled, further framing adjustments record the crop at the current position.
This is manual framing. Cutwise does not automatically follow a speaker or recognise which button matters. It gives me a way to make those decisions over time and check the complete result with Build preview.
There is also a Screen + webcam layout. It places the full composition in the upper 65% of the portrait frame and a chosen webcam region in the lower 35%. The webcam must already be part of the recording; this crops an existing picture rather than capturing or synchronising another camera file.
Shorts brand presets save caption styling, layout and the end message for reuse. Applying a preset preserves the current excerpt's timing and removes preset crop keyframes, so movement designed for one passage is not carried into another.
Making the finishing steps easier to judge
I added Duck music under speech to extra audio tracks. It lowers the music while the main audio is active and releases it during pauses, without changing the source recording or the main audio gain. The existing Original and Enhanced audio comparison remains available for judging the sound.
The export workspace now offers Main video, YouTube and Vertical short choices, with the relevant settings and a preview in one place. Vertical export uses a saved Short's timing, framing, captions and branding. Its file-size estimate is shown as a rough range because the final size depends on the video and encoder.
After rendering, Play export opens the result for checking and Show in Finder or Show in folder reveals the file. Resolution, quality and audio controls remain available under advanced settings, and main-video brand presets store output, audio and caption styling locally.
I also improved the recovery path for moved recordings. Locate video verifies the saved source hash when one is available and preserves the project's edits, captions and Shorts. Finding the same file in a new location should not be treated as replacing the recording with different footage.
Covering a region through the whole edit
The largest new finishing feature is a dedicated Effects track for blur, pixelation and solid covers.
On a paused frame, I can draw a rectangle, move it, resize its corners and set its start and end times. Preview effect renders the result before saving. The saved region becomes a timeline clip that can be moved, trimmed, split, duplicated, hidden or removed, with autosave and undo support.
The difficult part is keeping the cover attached to the footage as the edit changes. Regions follow their associated footage through cuts, inserts and splits, and appear in edited playback, MP4 exports and extracted PNG frames.
A moving transition introduces another problem. If a slide moves the sensitive pixels away from their original position before the cover is applied, a rectangle left at that position can expose them. I added protection to transition inputs before those pixels move, alongside the final composition layer above video, callouts and burned main-video captions.
New Shorts inherit the regions from their saved edit snapshot. The effects are applied before portrait framing, animated crops and the screen/webcam split, so those operations receive the already-covered picture. The framing canvas, preview and export use that same snapshot.
The implementation has specific regression tests for the first and last affected frames, moving transitions, caption layering and the different portrait layouts. Those checks address the places where a cover could appear correct in a still but fail during playback.
There are limits worth making explicit. Regions stay in fixed screen positions during their duration; moving or scrolling content needs the effect split and repositioned. Solid cover is the appropriate choice for passwords and API keys because blur and pixelation can leave readable detail. These visual effects do not remove spoken audio, transcript text or subtitle files, and new portrait captions are composed afterwards.
Saved Shorts also keep their existing snapshots. Adding a cover to the main edit does not silently change an older Short; a new Short is needed to include the new region.
Where this leaves Cutwise
This round made Cutwise more useful for finishing a recording: a clearer editing workspace, corrections that preserve transcript timing, direct selection of Shorts, moving portrait crops, music ducking, reusable presets and regional redaction.
The workspace implementation commit recorded 807 passing Flutter tests, three opt-in skips, clean static analysis and a successful Mac debug build. Later commits added the export refinements and redaction coverage. That recorded result belongs to the earlier workspace checkpoint; it is not a new qualification result for the final build.
Effects also move projects to schema 13. The previous project metadata is backed up during the upgrade, but older Cutwise builds cannot open the newer format. Original media stays unchanged.
The Windows Save As and accessibility work from the previous post still needs its remaining fixes and installed-application checks. These feature commits do not establish a new signed release or complete the separate clean-Mac and distribution work.
What they do provide is a more complete route through an edit: correct the words, review the cuts, frame the useful moments, cover what should not be visible and check the exported file. The next practical check is how well those pieces work together on real recordings.
