# Kinetic text behind a speaker: an editing guide

> Put kinetic text behind a talking-head speaker with Korpi Auto Edit. Learn how to choose footage, review text placement and adjust the finished edit.

Canonical page: https://korpi.ai/blog/kinetic-text-behind-speaker
Published: October 26, 2026 · 5 min read · Korpi AI Team

Kinetic text behind a speaker is a layered video effect: a graphic sits behind the person while the person remains visible in front. [Korpi Auto Edit](/tools/auto-edit) builds this treatment from a talking-head recording, using key lines from the transcript alongside punch-ins and captions.

The effect works best when the words reinforce the speaker's point. It is less useful when a long sentence fills the frame or a graphic arrives before the idea makes sense. Here is how to prepare a suitable clip and review the result.

## What the effect changes

The original background remains visible. During an emphasis moment, a kinetic graphic appears above that background, with the speaker layered in front of it. A matching punch-in emphasizes the same interval. Normal captions pause while the graphic is on screen, then return.

This is a visual emphasis treatment for an existing recording. It does not replace the speaker or record a new performance. For the complete upload and export sequence, see [the talking-head auto editing guide](/blog/talking-head-auto-editing).

## Choose footage with a readable silhouette

A stable camera, even lighting and a clear outline around the head and shoulders make the layering easier to read. Hair, fast-moving hands and a busy background are harder cases. Low-resolution footage can produce soft edges; expect to inspect the actual result rather than assume a studio-quality cutout.

Leave space around the speaker for a few short words. A tight close-up may hide most of the graphic. If a demonstration depends on objects near the edge, check that a vertical crop or punch-in preserves them.

## Make the edit in Auto Edit

1. Open [Auto Edit](/tools/auto-edit), then upload your talking-head clip or import a supported link.
2. Trim to a complete passage of up to two minutes. Choose the aspect ratio for your destination.
3. Review the estimate and choose Generate. Auto Edit transcribes the passage, selects key lines and prepares the layered graphic shots.
4. Wait for the final render, then play it all the way through. Check each graphic as it enters, holds and leaves.

Custom edits require an eligible paid plan and credits. The in-product estimate reflects the selected duration and your account's render pricing; [see the current plans](/pricing).

## Check meaning, placement and timing

Read the selected line against the full spoken sentence. A short phrase should preserve its meaning, especially if the speaker qualifies a claim immediately afterward. Reject emphasis that makes advice sound more absolute than it is.

Check the graphic at the narrowest crop. Can you read enough of the words around the person's silhouette? Is a large letter crossing the face? Are hair and hands clean enough for this source? Then watch the transition back to captions and listen for synchronization.

These checks are different from finding a highlight. If you are still choosing an excerpt, use [Auto Clipper](/create/auto-clipper) and its [context-preserving clipping checklist](/blog/how-to-make-auto-clipper-videos-with-ai) first.

## What you can change after preparation

Open the project in the editor to move or trim a graphic shot and adjust caption styling or zoom clips. The graphic words are baked into each rendered shot. Regenerate the edit to change those words; the editor does not expose them as editable text layers.

Keep a version with simpler emphasis when the speaker or background makes the effect hard to read. A clear, restrained edit is more useful than a graphic that hides the explanation. For more on choosing a workflow around existing footage, read [Korpi and OpusClip compared](/blog/korpi-ai-vs-opus-clip).

[Try Auto Edit with your own recording](/tools/auto-edit), then review the rendered result before publishing.