AI Video Editing: Rough Cuts and Subtitle Checks

For a first AI video-editing task, finish a talking-head rough cut and subtitles before adding effects. Keep the original footage, transcript, edit decisions and output separate so you can check what was removed and recover from a mistaken cut.

“Make this look good” is difficult to evaluate. You might want repeated sentences removed while the AI also removes a hesitation, a negation or a necessary pause. Start with a narrower goal: preserve the meaning, remove explicitly unwanted sections and align subtitles with the finished cut.

Choose footage you can evaluate easily

Record a one-minute practice clip introducing an object on your desk. Copy it into a test folder and preserve the original. Keep its aspect ratio for the first attempt so you do not have to solve reframing at the same time.

This article uses the official video-use workflow to explain the sequence from transcription to export. It does not report an end-to-end test of that tool. All timestamps and dialogue below are invented practice examples.

Check external services before installation

The upstream installation instructions require FFmpeg, Python dependencies and an ElevenLabs API key. Transcription uses an external service: open source does not mean every operation is free or that audio remains on your device throughout. Check available credits, processing terms and whether the footage can be uploaded.

Inspect an existing environment before reinstalling it. If you are not set up, ask the agent to read the official install.md, list missing components and potential costs, and install only what is necessary. Do not transcribe or upload footage before confirming those steps. Windows users should not copy macOS Homebrew or linking commands unchanged.

Write an editing brief with clear acceptance criteria

Source: the practice talking-head clip in the test folder. Do not overwrite it.
Goal: preserve the full meaning, remove retakes I explicitly mark, and add subtitles.
First: list the material, transcription requirements and external costs. Wait for my confirmation before transcription.
After transcription: propose keep/remove intervals with reasons before exporting.
Limits: no music, generated animation or changes in meaning. Ask me about uncertain cuts.
Deliverables: provide a rough cut first and retain edit decisions and subtitle files. Export the final version after review.

The brief separates transcription, edit decisions and rendering. You can approve them separately, and a subtitle correction should not require another analysis of unchanged footage.

Turn vague trimming instructions into edit decisions

Illustrative interval	Content	Suggested decision
00:00–00:04	Recording has started, but nobody is speaking.	Remove after confirming the visuals are unnecessary.
00:04–00:12	“You can adjust the height of this stand.”	Keep the complete sentence.
00:12–00:17	“But it cannot… cannot fold completely flat.”	Preserve the negation and limitation. Do not automatically delete every repeated word.
00:17–00:22	The speaker repeats the limitation.	Compare both versions and choose one complete statement.

This table is not an exported edit decision list or an executable editing instruction. It illustrates approval of editing intent. Actual cuts must use your footage and word timestamps; do not reuse these sample seconds.

A silent section may show a useful detail. Text alone can miss that value. Inspect the relevant visuals when needed, without sending every frame of the whole video to a model.

Align subtitles to the edited timeline

Correct names, product names and numbers in the transcript, then check timing against the edited video. Removing five seconds changes subsequent timing; the original timestamps cannot simply carry over. The upstream rules distinguish source time from output time.

Make a rough cut without extra animation. Check complete speech and transitions section by section. Once the subtitles pass, decide whether music, pictures or effects are useful. Each addition introduces another possible timing, volume or visibility problem.

Watch the entire export before accepting it

Compare the original and the cut for key points, negations, conditions and numbers.
Listen around every edit for clipped speech or abrupt audio.
Check subtitle readability, visibility and timing.
Open the export in the intended playback environment and check its aspect ratio, audio and file behavior.

For one incorrect subtitle, specify its location and replacement text while reusing the transcript and edit decisions. Upstream instructions require reusing transcription when the source is unchanged, avoiding repeated charges for the same transcription.

Understanding a video and producing an edited video are separate outcomes. Choose the workflow that matches the deliverable you actually need.

Sources and verification limits

The official README, installation guide and editing rules were checked on September 14, 2026. This article provides a method and invented examples. No transcription or paid-service call was performed, and no finished video or fixed percentage saving is claimed.