Ask ChatGPT to "remove the dead air from this Premiere sequence" and you'll get a script to paste in, or a description of which panel to open. You still do the work, and ChatGPT never sees whether the result is right. Ask the same thing to an agent that's actually holding a timeline, and the cuts happen, then the project lands in Premiere as an EDL. The difference isn't the model, it's what's on the other end of the sentence.
Can ChatGPT actually drive Premiere Pro?
Honestly, not on its own. ChatGPT in chat has no eyes on your project and no way to press a button inside Premiere's window. The paths people try to bridge that gap are the same ones that come up for any desktop NLE:
- The host's own scripting or extension tools, where they exist. Writing the script is something ChatGPT is genuinely good at; running it and judging whether it worked is still on you.
- Community MCP servers that wrap an editor's scripting layer so an agent can call it as a tool. A few of these exist for various NLEs; check what one actually exposes before trusting it with a real project.
- A plugin inside Premiere that calls an AI model for one task, captions or a text prompt. Useful for that task, but it isn't a timeline-aware agent, and it won't plan a multi-step edit.
None of these give ChatGPT a project it can look at, reason about and edit the way it reads and edits code. That's the gap "chatgpt premiere pro" searches are actually running into.
The route that gives it a real timeline
Shoty is a desktop video editor (Mac and Windows) with an agent chat built in. Pick ChatGPT in the model picker and the shape of the problem changes:
- Shoty runs the official Codex CLI on your machine, signed in with your own OpenAI account. It never sees your token and charges no fee for the agent.
- For each turn, Shoty opens a local MCP server and hands the agent tools: look at a frame, read and update the timeline, generate a shot, export. Every call shows up in the chat.
- The timeline is a text format (
.shoty), so the agent is reading and editing a structure instead of guessing at pixels or hoping a pasted script ran clean. - When the cut is done, Shoty exports it as EDL, FCPXML or OpenTimelineIO, the formats Premiere Pro, DaVinci Resolve or Final Cut Pro open for the finish.
The agent never touches Premiere directly. It edits a project Premiere can read afterward, which is the part the plugin-and-script route can't deliver.
1. Do the cut in Shoty, not in Premiere
Ask for the rough cut, the VFX fix or the subtitle pass while the project is still in Shoty.
Cut the dead air out of the first five minutes, keep room tone under the cuts, and add burned-in English subtitles.
2. Name what must not change
If a shot needs a fix before it leaves Shoty, say what stays locked so the model doesn't repaint the whole frame.
Replace the sky in shot 3 with an overcast afternoon. Keep the skyline, the camera move and everyone in frame exactly as they are.
3. Export for the host you're finishing in
Export this project as an EDL for Premiere Pro.
Ask for FCPXML instead for Final Cut Pro, or OpenTimelineIO if your next stop reads that format.
4. Finish where you already work
Color, mix and final delivery still happen in Premiere like they always have. Shoty's job ends at the cut and the AI shots.
Where this still needs you
- Picking the beat. The agent finds every pause fast; which one actually lands is still your call, and you can always grab the Shoty timeline and trim by hand.
- Reviewing generated shots. Anything sent to a video model costs credits, so set the run mode to review if you want to approve shots before they render.
- Round-trips. If notes from the Premiere side mean a shot needs another pass, that happens back in Shoty, not by scripting Premiere again.
Download the free Shoty editor and connect ChatGPT from the model picker. For the Claude side of the same setup, see Claude Code Premiere Pro.