Video Editor — Reframe, Jump-Cut, Captions, Overlays, And Animated Slides

August 19, 2026

Describe the cut and Alfrada OS can now do the cleanup pass that used to take a desktop editor: reframe a talking head to 9:16, strip the dead air, burn or export captions, overlay a lower-third, pull the soundtrack, and drop animated explainer slides between clips. Long renders move to a background job so the chat stays usable. You still finish the timeline in Live → Video Studio.

What you can do

• Ingest with a transcript, not just a probe. Ingest transcribes with word-level timing, labels speakers when there is more than one voice, lifts any embedded subtitle track, and describes the picture frame by frame. That is what makes "add captions" a one-line request. Window a long file — "ingest 12:00 to 14:30" — so you are not paying to analyse the whole recording. Anything over about twenty minutes returns a warning.
• Reframe and transform in one render. Crop to a target shape without letterboxing (anchor on the speaker), rotate and flip phone footage, speed / reverse / loop (true slow motion synthesises in-between frames), fade from or to black, set 24 / 25 / 30 / 50 / 60 fps, and clean up (stabilise, denoise, sharpen, deinterlace). One request, one re-encode.
• Overlay more than a corner watermark. Picture-in-picture, side-by-side or stacked comparison, a text card, or a timed lower-third. Overlay text is measured and wrapped so it does not run off the frame.
• Name a size the way you actually talk. Exact pixels (1080x1920), a platform preset (Instagram post / Reel, TikTok, YouTube Short or landscape, Twitter, LinkedIn, 2K cinema), or keep the source size so 4K stays 4K. Trims can start anywhere: "thirty seconds starting at 1:10".
• Hard cuts and breathing room. A zero-second transition is a hard cut. Ask for a pause — "half a second of black between each scene" — and the editor inserts it without touching the clips on either side.
• Colour beyond the twelve presets. Cinematic, noir, sketch, and the rest are a starting point. Describe a look they do not cover and Alfrada OS builds the grade; a custom look can layer on a preset.
• Captions three ways. Burn them in; export .srt or .vtt as a sidecar; or mux a real subtitle track the viewer can switch off. Word-by-word highlighting (the karaoke look) uses the word timings from ingest.
• Jump-cut a talking head. Silence removal strips dead air, keeps a little breathing room so words are not clipped, and reports how many seconds it actually took out. Scene split finds shot boundaries and can cut one file per shot.
• Lift the soundtrack. Extract audio to MP3, WAV, AAC, or FLAC. Narration stays locked to its picture across stitches; you can start a voice-over one second into a scene, freeze the last frame or add black to give a line room, and join per-line speech files with a set gap.
• Animated slides, not frozen cards. Title cards, bullet slides, full-frame image cards, and Manim slides (equations typeset properly, diagrams that build). Ask for a Ken Burns push on a still so a six-second photo does not read as a freeze-frame.

Where this shows up

You have a 12-minute demo recording and need a 45-second vertical cut for Reels. Ingest the window that matters, jump-cut the silences, reframe 9:16 around the speaker, burn karaoke captions, overlay the product name as a lower-third, finalize — then nudge the timeline in Video Studio.

You are building a narrated explainer from stills, a screen recording, and an equation. Image slides get a slow push, the Manim slide typesets the formula, narration is placed inside each scene so early lines leave a real pause, and music sits underneath without burying speech.

You restyled a trailer and do not want the chat frozen for five minutes. The render detaches; this conversation continues; the MP4 lands on the same card when it is done.

Try it

• "ingest in/demo.mp4 from 12:00 to 14:30, jump-cut the silences, reframe to 9:16 around the speaker, add karaoke captions, and make it an Instagram Reel"
• "overlay a lower-third that says Ada Chen, Head of Design from 0:03 to 0:08"
• "export captions as an srt I can hand to a translator, and also mux a switchable subtitle track into the MP4"
• "split this recording into one clip per shot, then I'll pick the strongest three"
• "extract the audio as wav, then rebuild the video with a Ken Burns still and a Manim slide of the unit-economics equation between the live clips"
• "restyle this trailer in watercolor, keep the faces and on-screen text, and don't freeze the chat while it renders"

Heads up

• Edits re-encode the whole file. Time scales with source length, not with how simple the ask sounds. Cut the window you need before a restyle or a grade.
• GIF output is a preview format. Capped at 640px wide and 12 fps, no audio, produced by the build step rather than finalize.
• Raw FFmpeg filter chains are allowed when a preset does not cover it. Filters that read files from disk are refused.
• Full Studio walkthrough: Live Video Studio. Capability reference: CUDA Video Editor.