How to Edit a Video Based on your Script with ChatGPT

A script tells you what a video needs to say. The recordings rarely arrive in that order: one instruction is spread across three takes, the best explanation begins with a false start, and the close-up you need is buried in another file.

With a compatible video editor connected through MCP, ChatGPT can help find those pieces and assemble them so the spoken instructions and visible actions work together.

You can use any editor with a compatible MCP connection, but if you’re looking for a free option, invideo Editor is a good choice. Once connected through MCP, ChatGPT can search your recordings, find suitable takes and assemble them around your script on a professional, frame-accurate multitrack timeline. You can also edit manually, shaping the cuts, framing, animation, color and sound as the video develops. To follow along with invideo Editor, start by connecting it to ChatGPT.

Connect ChatGPT to your editor

Open invideo’s connection page and choose ChatGPT. Follow the setup instructions and sign in with your invideo account when prompted.

Screenshot: the ChatGPT connection walkthrough on invideo’s Remote Agents page.

For the current custom-server route on ChatGPT web:

1. Open Plugins, select the plus button and choose Add custom MCP server.

2. Name the connection “invideo” and enter https://mcp.invideo.io/mcp as the server URL. Complete the invideo sign-in flow.

3. Create and install the resulting plugin, then select it in your editing conversation.

OpenAI’s custom MCP guide covers the current controls and workspace restrictions. Use the live connection instructions if your account presents a different setup flow.

Complete the connection before supplying the editing assignment. Then keep the project open in invideo Editor so you can watch the timeline, review selections and make your own changes. The complete ChatGPT editing guide covers the broader connection-to-export workflow.

Upload your script and footage and brief the agent

For this walkthrough, imagine you’re editing an espresso-machine setup tutorial for first-time owners. You have a script, several presenter takes and recorded demonstrations. Your goal is to assemble them into a clear, roughly three-minute video in 16:9.

Create a project called “First Espresso” in invideo Editor and upload your recordings. Share the project link with ChatGPT, then attach the script and machine notes in the same conversation.

Give ChatGPT the overall brief and ask it to place the recordings on the timeline:

Use the First Espresso project. We’re making a roughly three-minute, 16:9 espresso-machine setup tutorial for first-time owners. The video should make the setup easy to follow, using the presenter’s explanations and recorded demonstrations. First-Espresso-script.pdf contains the script; Machine-notes.pdf supplies the approved product details. Add all uploaded recordings to the “First Espresso v1” timeline, keeping the clips separate. Leave them untrimmed for now.

Illustration: the brief introduces the video’s purpose and asks ChatGPT to place the recordings on the timeline untrimmed.

This gives you a starting timeline with the source material in place. You don’t need to identify every usable take or timestamp yourself. The next instruction handles selection and assembly.

Assemble your footage to match the script

Once your clips are on the timeline, ask ChatGPT to assemble a base cut that follows your script. It should select the best takes and remove retakes, false starts, repeated lines and awkward pauses while keeping the delivery natural.

Follow First-Espresso-script.pdf and assemble the recordings into a coherent base cut on “First Espresso v1.” Choose the best complete takes and remove retakes, false starts, repeated lines and awkward pauses. Keep natural pacing, complete instructions and the script’s intended meaning. Aim for roughly three minutes without cutting necessary information. Flag anything required by the script that wasn’t recorded. Leave B-roll, graphics, subtitles and music for later.

Illustration: the selected takes form a base cut, with dialogue on a separate track and the other layers ready for later passes.

Play through the result with the script beside it. Check that the explanation follows the intended order, each line appears once and the cuts preserve complete words and natural delivery. Keep pauses that help viewers understand an instruction.

If a required line or action is missing, check the other recordings, record a pickup or approve a script change. Resolve that gap before refining the cut.

You can also adjust takes and cut points directly on the timeline. ChatGPT reads the updated timeline before continuing, so your manual decisions carry into the next pass. Once the base cut flows well, move on to B-roll and graphics.

Insert B-roll to support the instructions

Once the base cut flows well, ask ChatGPT to add relevant B-roll over the presenter’s narration. For this tutorial, use the recorded tank and portafilter close-ups to show the details being explained while keeping the dialogue continuous underneath.

In “First Espresso v1,” find the recorded tank and portafilter close-ups and place them on a video layer above the presenter during their matching explanations. Keep the approved dialogue continuous below. Use ranges that show each complete relevant action, with the hands and controls visible. Mute duplicate camera dialogue in the inserts. Preserve the base-cut duration, script wording and all other selections.

Illustration based on invideo Editor: the portafilter close-up sits above the base cut, while dialogue and recorded machine sound have separate audio layers. The example remains editable.

Watch each join with the preceding instruction. A shot can be relevant yet enter too late. Suppose the lock-and-brew passage works overall, but the close-up begins after the handle has already turned. The next instruction should target that selection without reopening the rest of the tutorial. The following times are illustrative; use the current timings in your edit.

Illustration: the instruction changes the close-up’s source in-point while preserving the passage duration, dialogue timing and approved sections.

Review the B-roll to check that each shot appears at the right moment and shows the complete action. Ask ChatGPT to adjust any shot that starts too late or misses an important detail. You can also change its timing or framing directly on the timeline.

Create step labels and graphics

With the footage in place, ask ChatGPT to add clear step labels and visual cues that make the tutorial easier to follow.

Add short, editable step labels to “First Espresso v1.” Use a consistent white-on-dark style and add simple arrows or callouts where they help explain the recorded action. Keep the hands and machine controls visible, use Machine-notes.pdf for approved terminology, and preserve the cut and dialogue timing.

Illustration: an editable step label and arrow callout sit above the recorded footage, with graphics kept on a separate timeline layer.

Play through the section to check that each label is easy to read and each callout appears with the action it explains. Ask ChatGPT to adjust anything that arrives too early, stays too long or covers an important detail. Once the graphics work with the footage, move on to subtitles and sound.

Add subtitles and refine the sound

Once the graphics are in place, ask ChatGPT to add clear, readable subtitles that match the spoken dialogue. Use the script and machine notes to check the wording and product terminology.

Add readable dialogue subtitles to “First Espresso v1.” Check the wording against the selected spoken takes and machine terminology against Machine-notes.pdf. Keep subtitles clear of the hands, machine controls and step labels. Preserve the approved picture, graphics and dialogue timing.

Then finish the sound as its own pass:

Balance the dialogue across the selected takes and smooth abrupt audio joins in “First Espresso v1.” Retain the real recorded machine sound during the matching brew action on a separate audio layer, without doubling the presenter’s voice. Add a quiet stock music bed, lowering it under instructions and fading it at the ending. Keep the approved picture, text and timings unchanged.

Illustration: subtitles accompany the dialogue, while recorded machine sound and music remain on separate audio layers.

Listen through the joins and the brew passage. If noise reduction thins the voice or music masks an instruction, ask for a lighter treatment at that point. You can adjust clip levels, fades and the mix with the Editor’s audio controls. The sound-effects and music guide develops separate layers, cue timing and mix refinement.

Review and export your video

Watch the finished video from start to finish, checking that it follows the script, the instructions are clear, and the visuals, subtitles and sound work together. Once you’re happy with the cut, ask ChatGPT to export that version.

Export the approved “First Espresso v1” timeline as a finished 16:9 H.264 MP4. Preserve the current picture, dialogue, subtitles, labels and sound mix. Keep the editable timeline available for revisions.

Illustration: choose H.264 MP4 for the finished video; the editable-project option remains available for a separate handoff.

Open the rendered file before sharing it and check picture, sound, text and the ending. A finished video and an editable project serve different purposes. If another editor will continue the work, request the relevant editable-project export for Premiere Pro, Final Cut Pro or DaVinci Resolve separately.

This approach fits a video whose message is already settled but whose recordings still need selection, assembly and refinement. You provide the script and creative decisions, ChatGPT handles repeated searches and timeline work, and the professional editing suite stays available whenever you want to shape the result yourself.

Frequently asked questions

Can ChatGPT assemble recordings to match a script?

Yes, through a compatible video editor connected over MCP. With the free invideo Editor MCP, it can inspect recordings, find suitable spoken takes and actions, and assemble them on an editable timeline. Supply the script, footage and any approved factual notes, then review how the selections convey each instruction.

Do I need to find the timestamps in my recordings first?

No. Describe the script beat or action and ask ChatGPT to find it. It can search the project footage and return source ranges. Review consequential choices, such as whether the selected take includes the complete instruction and its qualification.

What happens if a script line was never recorded?

Ask ChatGPT to flag it. Record a pickup, or approve a script change that matches the available material. You can also generate the missing line with a voice model and layer relevant B-roll over the voiceover. A missing visual action can sometimes be covered by another recorded angle; a generated illustration should remain distinct from footage documenting the actual procedure.

Does assembling a script-led edit use invideo credits?

Ordinary editing through invideo MCP, including footage analysis, cuts, subtitles and audio cleanup, does not consume invideo credits. Your ChatGPT subscription and usage allowance apply separately. Generating new images, video or audio uses invideo credits.

Can I change ChatGPT’s assembly manually?

Yes. Use the same professional timeline to change takes, trim, slip an insert, adjust framing, animate text or work on color and sound. ChatGPT reads the updated timeline before further edits. Keep the project link and a handover of approved decisions if you continue in a new conversation.