How to Apply Video Feedback with Claude or ChatGPT Without Rebuilding Your Edit

A new round of video feedback can mean reopening dozens of small decisions. You shorten one explanation, then move its close-up, retime the captions, and check that the music still lands where it should. The rest of the edit may already be approved.

Claude or ChatGPT can help carry out those revisions on an existing timeline. For this workflow, the assistant needs a compatible video editor connected through MCP. That connection gives it editing tools so it can change clips, graphics and audio from your instructions.

You can choose any editor with a compatible MCP connection, although setup and capabilities vary. If you want a free option, invideo Editor's MCP lets either assistant work on a professional, frame-accurate multitrack timeline with a full manual editing suite. You can direct the changes, make your own creative adjustments, and keep developing the same project.

Ordinary editing through invideo MCP does not consume invideo credits. Your Claude or ChatGPT plan and usage allowance apply separately. Generating new images, video or audio uses invideo credits.

The key is to give the assistant a revision brief with four things: the current edit, the exact target, the change you want, and what must stay. The prompts below work with either client.

Connect your assistant to the editing tools

Open invideo’s connection page and choose Claude or ChatGPT. Both connect to the same MCP server: https://mcp.invideo.io/mcp.

For Claude, open Customize → Connectors and add a custom connector named “invideo” with that server URL. Complete the connection and sign in with your invideo account. Enable the connector in the conversation you will use for editing. Managed accounts may need an organisation owner to add it first; Claude’s connector guide explains the current account-specific steps.

For ChatGPT, follow the connection page’s ChatGPT setup instructions. Add “invideo” as a custom MCP connection with the server URL above, sign in with your invideo account, and select it in your editing conversation. OpenAI’s connection guide covers the current setup; workspace permissions can affect availability.

invideo’s connection page shows the client-specific setup instructions and server URL. Choose the tab for Claude or ChatGPT.

Keep invideo Editor open alongside the conversation so you can play the revised edit and work on the timeline yourself.

Identify the edit that received the feedback

You can import an existing edit from Adobe Premiere Pro, DaVinci Resolve or another editor that exports a compatible XML project file, together with its source media. If you’re already working in invideo Editor, share the project link and name the timeline. If the reviewer watched an exported video, make sure you know which timeline version produced it. Feedback marked at 01:20 on an older cut may point to a different shot in the current one.

We’ll use an illustrative four-minute kitchen-makeover video for homeowners planning a small renovation. “Kitchen Makeover v2” already has a presenter-led sequence, room and detail shots, captions, music and a closing title. The opening before-and-after reveal and the layout explanation are approved. The latest notes ask for a shorter repeated explanation, a clearer drawer close-up and a closing title that is easier to read.

Give Claude or ChatGPT that context before requesting changes:

In the above illustration, the prompt identifies the current timeline and approved choices to preserve before requesting revisions.

The assistant can inspect the current footage, graphics, animation and audio through the MCP connection. Explain decisions it cannot infer from the timeline, such as why a particular take or pause has been approved.

If you want a comparison version, ask it to duplicate the timeline before the revision:

Duplicate “Kitchen Makeover v2” as “Kitchen Makeover v3.” Apply the feedback to v3 and leave v2 available for comparison.

This gives the review a clear starting point without assembling the video again from its source recordings.

Turn the feedback into one precise revision brief

Claude or ChatGPT can group related changes into a single pass. Replace vague notes such as “this part drags” with the repeated material to remove, the explanation to retain and the intended result.

Use timestamps from the reviewed edit, together with a spoken line, section name or shot description. That second reference helps identify the target after an earlier cut changes the timing. If two reviewers disagree about the same moment, resolve the desired treatment before asking the assistant to execute both notes.

Here is one coherent brief for our example. The ranges are illustrative; substitute the timings in your own edit:

Apply these notes to “Kitchen Makeover v3.” All timestamps refer to the starting v2 review cut. Identify all three targets before changing its timing.

1. At 01:10–01:22, remove the second explanation of why we chose deep drawers. Keep the first complete explanation immediately before it and the following demonstration of the drawer opening. Close the gap and carry the later clips and their attached captions/graphics with the edit. Preserve the words and shot choices in the approved layout section; its start time can move earlier.

2. At the original 02:05–02:14, during “Everything is easier to reach,” replace only the upper B-roll shot with the uploaded close-up showing the drawer fully opening. Keep the presenter’s dialogue and this passage’s duration unchanged. Show the complete opening action and return to the presenter at the existing cut.

3. On the closing title “Plan around how you cook,” keep the wording, placement and entrance animation. Extend the readable hold by one second using the available room shot, then move the title’s exit and music fade to the new ending. Leave the presenter’s final sentence unchanged.

Keep the opening reveal, the approved layout explanation and all other editorial choices unchanged. Return a short summary of the changes and their locations on v3, noting anything you could not complete.

See BMF MEDIA’s opening-revision notes beside the edited timeline at 04:50.

BMF MEDIA’s revised timeline and notes about tightening the opening.

The first note changes the sequence’s duration. The second changes picture over existing dialogue. The third extends a specific graphic and the ending around it. Those are different operations, so the brief specifies what should happen to the surrounding layers instead of asking the assistant to “make everything tighter.”

Use references to define the part you want changed

Claude or ChatGPT can also use a reference to guide a revision. Specify the relevant moment and the treatment you want carried over, such as the framing, title position or animation. The reference should narrow the request.

For example, suppose the drawer close-up now shows the action clearly, but the reviewer wants its label placed like one in a reference video:

In the above illustration, the attached reference guides label placement while the shot, dialogue and subtitle timing stay unchanged.

A screenshot is useful for a static layout; a video is better when the requested change concerns motion or pacing. Identify what the reference should influence rather than asking for its entire style. For more on developing an editable treatment, see creating custom video effects from a description or reference.

Review the changed passages and their joins

When the assistant finishes, use its change summary to locate the revisions. Play each passage with a little of the surrounding sequence, then watch the full edit. A completion message helps you navigate; playback is where you judge the result.

For our example, check that the retained drawer explanation remains complete, the close-up shows the action while the matching line plays, and the closing title stays readable without extending the ending awkwardly. Listen at the removed explanation’s join and check the captions and graphics that moved with it. Revisit the approved opening and layout section too.

Illustration based on invideo Editor: review the drawer close-up on V2 with its label above, presenter footage on V1, dialogue on A1 and music on A2. The playhead and preview show the same revised passage.

If one part needs another adjustment, use the current v3 timing or the named action rather than repeating the old review timestamps:

In v3, the drawer close-up returns to the presenter before the drawer is fully open. Use a source range that includes the complete opening action within the same passage. Keep the dialogue, passage duration and all other revisions unchanged.

You can also reverse one decision while keeping the others:

Restore the previous upper B-roll shot during “Everything is easier to reach.” Keep the removed repeated explanation and the longer closing-title hold.

invideo Editor supports undo and restoring earlier checkpoints. A targeted reversal is useful when you want to retain the other changes; comparing with v2 helps you decide which treatment works best.

Continue from your manual choices

You can make a precise creative adjustment yourself on the same timeline. Choose the exact cut frame, reposition a label, shape a fade or refine an animation with the Editor’s manual controls. Claude and ChatGPT reread the current timeline before continuing, so the next task can build on those changes.

Explain the intention when it should guide the next pass. For example, if you manually give the kitchen reveal more breathing room:

Continue from the current v3 timeline. Keep the opening reveal and its new hold as they are; the extra beat lets viewers compare the room. Match the “Storage” section-title animation to the approved “Layout” title, keeping its wording and section start unchanged. Leave the picture edit and dialogue untouched.

For a separate example, Luuk Alleman requests a new opening after manual edits at 04:10. The demonstration shows the opening request and playback on the editable timeline.

Luuk Alleman’s new-opening notes refer to his manual edits; playback remains available on the same timeline.

If you switch clients or start a new conversation, bring the project link, timeline name, approved choices and remaining work. The new assistant can read the edit, but the previous discussion does not automatically transfer. Ask the current conversation to prepare that handover before switching.

Export the approved edit

Once the revisions work, identify the approved timeline and ask Claude or ChatGPT to export it as a finished H.264 MP4. If you need to continue in another editing application, request an editable-project export separately. Open the rendered file and check the changed passages, sound and ending before sharing it for the next review.

The export options show H.264 MP4 selected for Kitchen Makeover v3, with the revised drawer close-up in the preview.

This workflow suits editors who want to delegate repeated revision work while keeping control of the creative decisions. The edit stays available for further direction and manual refinement as feedback develops.

Frequently asked questions

Can Claude or ChatGPT apply feedback to an existing edit?

Yes, through a compatible connected editor. In invideo Editor, share the project link and timeline name, then identify the changes and approved elements to preserve. The assistant works from the current timeline.

Do I need a timestamp for every change?

No. A section name, spoken line or clear clip description can identify a target. Use current-edit timestamps for precise changes, and pair them with a content reference when earlier cuts may shift the timing.

Will shortening one section move everything after it?

It can. State whether you want to close the gap and move later material earlier, or keep the passage’s duration. When timing changes, check the related captions, graphics, music and any chapter timestamps against the new edit.

Can I undo one revision without losing the others?

You can ask the assistant to reverse a specific change while retaining the rest. invideo Editor also supports undo and earlier checkpoints. Keep a separate timeline when you want to compare an alternative with an approved version.

Is applying feedback through MCP free?

Ordinary editing through invideo MCP uses no invideo credits. Your Claude or ChatGPT plan and usage allowance apply separately. Generating new images, video or audio uses invideo credits.