Video Editing Prompts for Claude and ChatGPT: What to Ask at Each Stage of the edit

Finding a usable take, assembling a base cut and lining up a close-up or B-roll with the right words can mean a lot of scrubbing and small timeline adjustments. Asking Claude or ChatGPT to “make it better” leaves those decisions open. A useful editing prompt tells it which job to carry out, what a successful result should show or sound like, and what you want to preserve.

This guide explains how to prompt your AI assistant for each major stage of your editing process. The prompts below work with both Claude and ChatGPT. Both of these tools however need an active connection to a video editor through MCP to access its controls and carry out these changes.

We’ll use invideo Editor’s free MCP connection for the examples. You can use another editor that offers a compatible MCP connection, although its capabilities and setup may differ.

All the work we do with these will stay fully accessible and editable on invideo Editor’s timeline, with manual editing controls. Claude or ChatGPT can continue from the changes you make yourself.

How to connect ChatGPT or Claude to your editor

For invideo Editor, open invideo’s connection page and choose your client. Both use the MCP server URL https://mcp.invideo.io/mcp.

For Claude, open Customize → Connectors → + Add → Add custom connector. Name it “invideo,” enter the server URL, complete the connection and sign in with your invideo account. Enable it for the editing conversation. Team and Enterprise accounts may need an organization owner to add the connector first. Claude’s custom-connector guide explains the current account-specific route.

The Claude tab on invideo’s connection page shows the custom-connector route and server URL.

For ChatGPT, open ChatGPT Plugins, select + and choose Add custom MCP server. Name it “invideo,” enter the URL, configure its sign-in and create the plugin. Install it, then select it in your editing conversation. Follow invideo’s connection page if your account requires developer mode; workspace permissions can affect availability. OpenAI’s connection instructions cover the current plugin setup.

The ChatGPT tab shows its setup route. Enable the connection in the conversation you’ll use for editing.

What to ask at each stage of your edit

Each stage calls for a different kind of instruction. Use the table to identify the task, describe the result you want and understand what the assistant should produce.

Detailed prompts for each editing stage

The prompts below are for a guide/tutorial built from recorded explanations, demonstrations and supporting footage. Feel free to adapt it to whatever you are making.

Replace the text in [square brackets] with your topic, audience, source files, timeline name and delivery requirements. Use the same project and timeline references across prompts, and remove fields that do not apply.

1. Uploading footage and giving the project context

An upload prompt should say where the files belong and what each recording contributes. The editing brief adds the audience and purpose, so later requests have something specific to work toward.

Create a project named [project name] and import [attached files or accessible file paths]. The footage includes [presenter recordings], [close-ups of the demonstrated action] and [before-and-after shots]. We’re making a [target duration], [aspect ratio] guide about [topic] for [audience]. Keep the presenter’s complete explanation of [essential point]. Confirm which files are available and their roles before changing a timeline.

Illustration: presenter recordings, demonstrations and supporting shots are organized as source files for the tutorial.

Explain which recordings carry the instructions, which show the actions and which establish the result. Naming those roles is more useful than uploading a folder and asking for a video.

ChatGPT can import attached files into the connected project. Claude Desktop can upload local footage; with Claude on the web, upload the files directly in invideo Editor. If the media is already available, replace the upload request with:

Open [project link] and read the available footage. Work on [timeline name]. This video is for [audience] and should help them [intended takeaway]. Preserve [approved choices or manual changes]. Confirm the source material and current edit before making changes.

What you should get: the recordings in the correct project and a shared understanding of the video. This first prompt establishes context; the next one requests the cut.

2. Assembling a base cut

Explain the order of the ideas and which material carries them. A target duration helps, but the assistant also needs to know which explanations and actions must survive the shortening.

Create [timeline name] as a [aspect ratio] timeline. Open with [opening shot] and the recorded line about [opening point]. Assemble the explanation of [first step], [second step] and [result], in that order. Remove abandoned takes, repeated explanations and idle waits. Keep [essential qualification] and pauses needed to show [demonstrated action]. Aim for [target duration] without cutting an essential instruction. Leave B-roll, graphics and music for later passes.

Illustration: the assembled timeline with the opening, steps and result arranged as individual clips, with recorded dialogue on its own track.

Name the essential steps and explain which actions need time on screen. This gives the assistant a reason to retain a quiet passage when it helps the viewer follow the demonstration.

What you should get: a base sequence with separate clips and cut points. Watch whether the opening leads into the first step and whether the ending delivers what the introduction promised.

If you have not chosen an opening yet, ask for options before assembly:

Find [number] opening options that show [problem or result] and pair each with a complete recorded line about [topic]. Return filenames and source timestamps so I can watch them. Leave the timeline unchanged until I choose.

This gives you a selection to judge rather than an opening chosen from an unclear brief.

3. Adding B-roll to the edit

A B-roll prompt needs both a shot description and a placement instruction. Say which source to use, what the viewer should see, and what happens to the dialogue underneath.

In [timeline name], find an uploaded shot showing [specific action or detail]. Place it above the presenter footage during the line “[spoken line]” at [current timeline range, if known]. Show the complete action, mute the insert’s source audio and keep the presenter dialogue underneath. Return to the presenter before [next line or action]. Preserve the passage’s duration and frame [important details] together.

Illustration: a muted close-up sits above the presenter footage while the recorded dialogue continues underneath.

Describe the complete action and the details that must remain visible. Footage search lets you describe the shot even when you do not know its filename.

What you should get: an insert timed to the explanation, while the base cut and dialogue continue underneath. Check that the selected source includes the whole movement.

Specify the sourcing route when your own recordings do not contain the shot:

For stock footage:

Source a stock shot of [subject or action] for [passage or purpose]. Match [framing, setting or style]. Place it during [line or timeline range], preserving [dialogue and timing]. Use stock footage for this insert.

A request to generate a shot is different from searching uploaded footage or sourcing stock. Use the actual recordings to show the demonstrated technique and result; a generated diagram can explain an idea.

To generate a new supporting shot:

Generate a [duration] B-roll clip showing [subject or action] to support the explanation of [idea] in [timeline name]. Use [visual style], [setting and lighting], [framing and camera movement] and [aspect ratio]. Place it during “[spoken line]” at [current timeline start–end timestamps]. Mute the generated clip’s audio and keep the presenter dialogue underneath. Preserve the passage’s duration, captions, labels and other shots. Use this as an illustrative insert; keep the recorded demonstration and result.

4. Inserting graphics and captions

Give exact wording and describe where it can sit without covering the action. Separate a step label’s job from the spoken captions’ job.

On [timeline name], add an editable step label reading “[exact label text]” during [action or current timeline range]. Use [font or reference style], [size] and [colors]. Place it in [clear area of the frame], away from [important subject or detail]. Keep it visible from [start cue] to [end cue]. Add spoken captions in [caption style and position], clear of the label and action. Preserve the picture and dialogue timing.

Illustration: step labels and spoken captions occupy separate layers above the demonstration footage.

Use step labels to orient the viewer and captions to carry the spoken explanation. Keep both clear of the action being demonstrated.

What you should get: editable labels and captions with their own placement and timing. Review long lines and moments where hands or tools move into the text area.

For a targeted refinement:

Keep the caption wording, font and timing. Move captions in [passage] to [new position] so they stay clear of [hands, product or face]. Leave all other captions, labels and footage unchanged.

5. Building effects and transitions

Describe the visual purpose before naming a treatment. Identify the shot, graphic or join it affects, define its timing and specify the controls you want for later adjustments.

On [timeline name], animate the “[label text]” layer with an opacity fade from [start cue] to [fully visible cue]. Hold it through [action] and fade it out before [next step]. Keep its position, wording, font and size unchanged. Apply the same entrance to [other named labels], preserving each label’s individual timing. Leave captions and footage unchanged.

Illustration: the step label is partially transparent in the monitor, with its opacity keyframes open on the right and its text clip selected above the footage.

Keyframe controls keep the entrance, hold and exit editable. Name the properties you want to adjust, such as opacity, position, scale or timing.

For a footage effect, describe the treatment and the controls it should expose:

On [timeline name], build [effect or visual treatment] on [clip or current timeline timestamps] to create [intended visual result]. Make [properties I want to adjust] editable, with controls for [range or behavior]. Preserve [picture timing, dialogue or other elements]. Leave other clips unchanged.

For a transition, identify the exact join:

At the join between [shot A] and [shot B], add [transition type] lasting [duration] to create [intended feeling or connection]. Keep controls for [transition properties I want to adjust] editable. Preserve [important action, dialogue or total timing]. Leave other joins unchanged.

What you should get: the requested treatment on the named layer, shot or join. If its timing competes with the action, revise that timing rather than replacing the whole visual style:

Keep [effect or transition] and its current settings. Start it at [new cue] and finish it by [new cue]. Preserve [elements that should stay].

6. Making precise edits to sections

Identify the passage with its current timeline start and end timestamps, plus a line or shot description. Then say whether the change should preserve its duration or shorten the sequence.

For a picture-only replacement:

In [timeline name], replace only the insert during “[spoken line or step label]” at [current timeline start–end timestamps]. Choose the uploaded take showing [complete action]. Fit it within the existing [passage duration] without speeding up the action. Keep the presenter dialogue, passage duration, captions, labels, music and other shots unchanged. Return to the presenter at the existing cut. If no take fits, show me the alternatives before changing the timing.

Illustration: a wide shot is replaced by an overhead close-up on V2, while the 00:20–00:32 boundaries, presenter footage and dialogue remain unchanged.

For a removal that should shorten the edit:

In [timeline name], remove the repeated explanation of [topic] at [current timeline start–end timestamps]. Keep [first complete explanation] and [following instruction]. Close the gap and move later clips, captions and labels earlier with the edit. Preserve their wording and shot choices. Retain [deliberate pause or reaction] and adjust the music at the join without changing its treatment elsewhere.

What you should get: either a different picture within the same passage or a shorter sequence with later material moving together. These are different editing decisions; “fix this section” does not explain which one you intend.

After a duration change, use the current timeline’s timings for the next note. To reverse one choice without undoing the rest:

Restore [previous shot or treatment] during [line or passage]. Keep [approved removal or revision] in place. Preserve every other change.

7. Sound and music

Specify what should lead the mix and which sounds matter. Name the music source and describe its placement, rather than asking for a “professional soundtrack.”

On [timeline name], balance [dialogue track or speaker] and reduce [distracting noise], keeping the voice natural. Add [uploaded track name or requested stock music style] beneath [passages]. Keep it quiet under speech, lower it during [important action or sound], and fade it in at [start cue] and out at [end cue]. Keep [specific recorded sound] audible. Preserve all picture cuts, captions and labels.

Illustration: dialogue, music and an action sound sit on separate tracks; the music lowers during the highlighted passage.

Name any recorded action sound that should remain audible alongside the explanation. Its source audio needs to be included in the edit.

What you should get: dialogue, music and effects placed and balanced across their tracks. Listen at the joins and during the demonstration. You can continue adjusting levels, fades and track balance manually.

If the music still competes with the voice:

Lower only the music beneath [line or current timeline range]. Keep the chosen track, its placement, the dialogue level and the rest of the mix unchanged.

8. Resizing and making cutdowns

A resize keeps the content but adapts its framing. A cutdown chooses a shorter story. Tell the assistant which job you want, and preserve the master on a separate timeline.

For resizing:

Create a separate [new aspect ratio] version of [source timeline name], named [new timeline name]. Keep its content and duration. Reframe around [subject and important details], reposition captions and labels for the new canvas, and keep text within [destination’s safe areas or supplied reference]. Leave the source timeline unchanged.

Illustration: the full guide’s timeline is shown on the left and a separate portrait cutdown on the right, retaining the opening, first step and result.

For a cutdown:

Create a separate [aspect ratio] timeline named [new timeline name] from [source timeline name]. Make a roughly [target duration] video for [destination and audience] that communicates [one takeaway]. Include [opening problem], [essential action or explanation] and [result]. Use the presenter’s actual recorded words and retain [necessary qualification]. Reframe around [subject], adapt captions to the new format and leave the full guide unchanged.

Define whether the cutdown should introduce the topic or teach a complete process. Preserve the explanations and qualifications needed for that purpose.

What you should get: a separate version with framing and text adapted to its purpose. Watch it on its own to check whether its meaning survives the shorter selection. Our ChatGPT Shorts guide develops cutdown selection in more detail.

9. Exporting the edit

Name the version to deliver and the kind of file you need. A finished video and an editable project for another application are separate outputs.

For a finished video:

Export the current [timeline name] as a finished [aspect ratio] H.264 MP4 for [destination]. Use [requested resolution and frame rate]. Export this version only, with its current picture, captions and sound.

Illustration: MP4 is selected in Export Settings, with the format, resolution and frame-rate options beside the project-format choices and the tutorial timeline visible behind the panel.

For an editable project:

Export [timeline name] as an editable project for [Premiere Pro, Final Cut Pro or DaVinci Resolve], including the timeline and media used in the edit.

What you should get: the requested deliverable from the named version. invideo Editor’s save-and-export options include finished video and editable-project delivery. Open the delivered file or project to check its content, sound and ending before sharing it.

Across these stages, the useful prompt names the task, its target, the intended result and what should stay. Give the overall brief once, then direct the work your current edit needs. Keep invideo Editor open alongside the conversation so you can play the changes and continue shaping cuts, layers, graphics and sound with its full manual editing suite.

Frequently asked questions

Do these prompts work with both Claude and ChatGPT?

Yes, when the assistant has an active MCP connection to a compatible video editor. The task, target, intended result and preservation rules apply to either client. Adapt the prompt to the controls your chosen editor exposes.

What should I put in the placeholders?

Replace each bracketed field with your project’s details: its link, timeline name, audience, desired duration, spoken line, shot description or visual style. Remove fields that do not apply. Do not leave conflicting instructions from the example in your prompt.

Do I need exact timestamps for every request?

No. Describe the spoken content or visual action and ask the assistant to find matching footage. For precise revisions, use a current timeline range plus a line or shot description. Source timestamps and positions on the edited timeline are different references.

Should I use one prompt for the entire edit?

Give the overall brief first, then request the stages your edit needs. Combine related actions when their intended result is clear, such as finding an insert and placing it over the matching explanation. Ask for options before execution when you still need to choose.

Does editing through invideo MCP use credits?

Ordinary editing through invideo MCP does not consume invideo credits. Your external assistant’s plan and usage allowance apply separately. Stock sourcing by the invideo agent and generating new images, video or audio use invideo credits.