# Edit video by chatting

Canonical URL: https://getvidiyo.app/features/chat-video-editing
Last updated: 2026-10-08

> TL;DR: In Vidiyo you edit by typing requests like "cut the ums and long pauses" into a chat panel. The assistant reads the word-timed transcript, looks at rendered frames, makes the change on a real timeline and checks its own work. Every request undoes in one step, and you can finish anything by hand.

Image: Vidiyo editor with the AI panel open, showing a plan card waiting for approval next to a vertical talking-head video and its timeline.

Chat editing in Vidiyo means you describe the edit in plain words and the assistant does it on a real timeline, not in a black box. You see every cut, caption and layer it adds, you can undo the whole request in one step, and you can take over by hand at any point.

## How do you edit a video by chatting in Vidiyo?

You open a project, import your footage and type a request into the AI panel on the right. The assistant reads the transcript, makes the edit, checks the result and replies with what changed. A short preview of the result renders under its reply, so you can watch it before moving on.

Things people type:

- "Cut the ums and long pauses."
- "Start on the strongest line."
- "Put a counter on \"160 GB\"."
- "Show the product above me."
- "Cover the jump cut with B-roll."
- "Keep the music under my voice."
- "Warm up the color."
- "Review it before I export."

While it works you can send another message to steer it, and when it finishes you can press Tab to use its suggested next request. You can also attach up to eight files to a request: images as visual references, text, SRT or Markdown as reference content, and video or audio, which gets imported into the project.

## What does the AI see when it edits?

It sees more than the words. The assistant gets a readable edit list of the timeline, the word-timed transcript, measured face positions and loudness, and rendered frames it asks for. For a wider look it requests a contact sheet: composed frames from risky moments, labelled with timecodes and outlined faces, caption boxes and on-screen text.

| What it reads | Why it matters |
| --- | --- |
| Word-timed transcript | Cuts land on word boundaries, and graphics anchor to the words you say |
| Rendered frames | It can see a caption over your face or a title off the edge |
| Measured loudness | Music sits under your voice at a measured level, not a guess |
| Quality checks | Dead air, raw jump cuts and cropped faces get flagged after every edit |

This is the difference between a frame-aware editor and a [transcript-based editor](https://getvidiyo.app/glossary/transcript-based-editing). A transcript can tell an AI what you said. Only the frames tell it what the viewer sees.

## How does it work under the hood?

The assistant edits through the same validated edit engine as the editor itself. It doesn't write files or replace your project. It sends structured edits, like "delete these words" or "tighten pauses longer than half a second", and Vidiyo computes the exact frames from measurements, such as where speech actually stops.

After each batch of edits, Vidiyo runs quality checks on that change and hands the findings back to the assistant. Before export it can run a full check: captions over faces or under the platform interface, frames with no picture, slow openings, uncaptioned speech and loudness against the -14 LUFS social target. A separate critic review, which can't change anything, looks at the cut with fresh eyes and pins its notes as comments on the timeline.

For big jobs like a first cut or a restructure, you can switch to plan mode. The assistant shows a plan card with ordered steps and two lists: what you asked for and what it assumed (length, tone, which take). You correct the assumptions, then approve.

It also remembers things. Settings → Memory shows what it keeps about you and about each project, and you can edit or delete it.

## What can't chat editing do?

Chat can only do what the editor can do. There's no speed control, so it can't make a timelapse with a speed ramp (it can build one from jump cuts). Zooms are keyframed punch-ins, not automatic cursor-follow. Timelines stop at 5 minutes and export at 1080p, 30 fps. Colour tools are basic.

The assistant also can't approve or publish a post for you. It can draft captions and schedule details, but sending to Buffer always needs your explicit approval.

## Who is chat editing for?

It's for people who record themselves and don't want to spend an evening on a timeline. Founders making launch and demo videos, creators posting talking-head shorts, and agencies turning around client edits get the most from it. If you already pay for Claude or ChatGPT, there's no second AI subscription to add. Vidiyo itself is free during its invite-only beta.

If you'd rather edit by hand, you can. The [layers and keyframes](https://getvidiyo.app/features/layers-and-keyframes) are there, and the assistant's changes are ordinary layers you can grab and move.

## Honest limits

- Timelines max out at 5 minutes, and export is 1080p at 30 fps.
- Chat needs Claude Code or Codex installed and signed in. Editing and export work without it.
- No speed ramps or speed control, so the assistant can't speed up or slow down a clip.
- Every request counts toward your Claude or ChatGPT plan's usage limits.

## Frequently asked questions

### Can I edit a video just by typing what I want?

Yes. In Vidiyo you type a request such as "start on the strongest line" and the assistant makes the edit on the timeline. Each request becomes one undo step, and every change stays editable by hand.

### Does Vidiyo's AI actually watch the video?

It reads the word-timed transcript and also looks at rendered frames it picks, plus contact sheets marked with faces, captions and on-screen text. That is how it notices captions covering a face or a frozen shot, which a transcript alone can't show.

### Which AI does Vidiyo use for chat editing?

It uses the plan you already have: Claude Pro or Max through Claude Code, or a ChatGPT plan through Codex. Vidiyo has no AI subscription of its own, and usage counts toward your plan's limits.

### Can I stop the AI from making changes I didn't ask for?

Yes. Ask for a plan first and the assistant shows a plan card that separates what you asked for from what it assumed. Nothing changes until you approve. You can also undo any request in one step.

### Can I still edit by hand?

Yes. Vidiyo is a full timeline editor with layers, keyframes, captions and audio tracks. The assistant uses the same tools you do, and a manual edit stops whatever request is running.

## Related

- [Remove silences and filler words automatically](https://getvidiyo.app/features/silence-and-filler-word-removal)
- [Comment on the frame, then Fix with AI](https://getvidiyo.app/features/comments-fix-with-ai)
- [Can Claude edit videos?](https://getvidiyo.app/claude-video-editor)
- [Can ChatGPT edit my video?](https://getvidiyo.app/chatgpt-video-editor)
- [Agentic video editing](https://getvidiyo.app/glossary/agentic-video-editing)
- [Transcript-based vs frame-aware AI video editing](https://getvidiyo.app/blog/transcript-based-vs-frame-aware-ai-video-editing)

Request access to the Vidiyo beta: https://getvidiyo.app/request