How GPT-Video's Vibe Editing Works

GPT-Video's vibe editing replaces the timeline with a text prompt: you describe the change you want, and an Editor Assistant plans the edit, applies it to your footage, and lets you undo any step. It fits creators, editors, and clippers who already have raw footage and want clips cut, reframed, graded, and titled without dragging clips around a timeline. You still need source video and a project to edit — the tool directs the edit, it doesn't shoot or source footage for you.

The core idea: describe the edit, not the timeline

The site's framing is that "ChatGPT changed how you write. GPT-Video changes how you edit." Instead of keyframes and tracks, you write instructions in plain language. The Editor Assistant reads the project state, decides what operations to run, and executes them on your footage.

The page describes this as "No timeline, no keyframes. Write what you want changed, and the Editor Assistant plans the edit, runs it on your footage and lets you undo any step."

So the loop is:

1. You write a description of the desired change.
2. The Editor Assistant plans the edit — it interprets your instruction against the current project.
3. It runs the edit on your footage.
4. You review and undo any step that missed.

What a prompt actually looks like

The site shows a worked example: a night-stunts project with four shots and a 0:12 runtime. The user's instructions include things like:

• "Cut the best 3 seconds of each take into one reel"
• "Make it vertical for TikTok and keep the rider in frame"
• "Make the colours pop for the feed"
• "Push in slowly on the burnout"
• "Put WAIT FOR THE SMOKE on the first two seconds"
• "Dip to black between the night shots and the sunset"

The assistant responds with what it did, for example: "Added a 0.4 s dip to black at 0:09, where the night burnout cuts to the sunset wheelie." That response is the key mechanic — the edit is reported back in concrete terms (operation, duration, timestamp, and the shots involved), so you can judge whether it matched your intent before moving on.

The edit types it handles

Based on the example and the listed capabilities, vibe editing covers:

Formats it produces natively

The site lists viral formats as built-in targets rather than things you assemble manually: podcast → 10 clips, fake text story, Reddit story video, split screen, stream highlights, and animated captions. If your source is a long podcast or stream, these are the output shapes the tool is oriented around.

Why "undo any step" matters

Because the assistant acts on your footage in sequence, a wrong interpretation early can cascade. The page states every edit can be undone, which makes the workflow iterative: describe, check the reported result, undo or refine, then describe the next change. Treat the assistant's confirmation message as your verification step — it tells you what changed and where, so you don't have to scrub the whole clip to find out.

Three steps to posted

The page's own framing is "Three steps to posted," which maps to the loop above: describe the edit, let the assistant run it, then export (including vertical 9:16 for short-form platforms).

When this approach fits

Vibe editing is a good match if you have long footage and want many short clips, you think in outcomes ("make it vertical," "cut the best 3 seconds") rather than frame numbers, and you're willing to iterate on prompts. It's a weaker match if you need frame-exact manual control on every cut, or if you don't yet have source footage to feed it.

For current plan details and any usage limits, check the site's Pricing and API pricing pages directly — the page evidence here doesn't specify what's free or what requires an account.