How to Find, Generate and Add B-Roll to Your Edit with ChatGPT

Your main edit is working, but adding B-roll can mean another long round of scrubbing through recordings, comparing close-ups and hunting for shots you never filmed. Then comes the fiddly work: choosing the useful part of each shot, lining it up with the right words and trimming it so viewers can follow the action. If a shot arrives too early or ends before the demonstration is clear, you’re back on the timeline adjusting it again, while trying to keep the dialogue and approved edit intact.

ChatGPT can help with that work when connected to a compatible video editor through MCP. The connection gives it editing tools to search footage, source or generate supporting visuals, and arrange them on your timeline. A conversation alone does not reliably carry out this editing workflow.

You can choose any editor with a compatible MCP connection; setup and available capabilities vary. If you want a free option, invideo Editor’s MCP connection lets ChatGPT work on a professional-grade, frame-accurate multitrack timeline. The full editing suite remains available manually, so you can direct the B-roll pass in chat and make your own creative adjustments to the same edit.

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

In this guide, we'll use an existing six-minute pour-over coffee tutorial as our worked example. The presenter edit, dialogue and music are already in place. The project also contains overhead brewing footage and close-ups of grinding, pouring and the coffee bed. Our job is to make the explanation easier to follow with a few well-chosen visuals. The prompts and timings below are suggested examples; replace them with your own project's details.

Connect ChatGPT to invideo Editor

Open invideo’s connection page, choose ChatGPT and follow the instructions for your account. On the web, open ChatGPT Plugins, select + and add a custom MCP server named “invideo” with this URL:

https://mcp.invideo.io/mcp

Sign in to invideo when prompted, install the connection and select it in your editing conversation. Follow the connection page if your account requires developer mode or workspace permission.

The ChatGPT tab on invideo’s connection page shows the MCP server URL and setup route.

Keep invideo Editor open alongside ChatGPT so you can play the timeline and make manual adjustments as the edit develops.

Choose your edit and explain what the B-roll should clarify

Give ChatGPT the existing project link and the timeline name. You can upload additional recordings in invideo Editor or attach them in ChatGPT and ask it to add them to the project. If your edit is in Premiere Pro or DaVinci Resolve, import its XML project file and source media into invideo Editor to continue from an editable timeline.

For our example, start with the context and boundaries:

The illustration shows a brief that identifies the timeline, the visuals needed and the approved parts of the edit to preserve.

Find supporting shots in your footage

ChatGPT can search the connected project's footage by visual action or spoken content. Describe what the viewer needs to see and the line it should accompany, then ask it to place the relevant selection. You do not need to identify the filename or watch every recording first.

In our tutorial, the presenter says, “Pour just enough water to wet all the grounds, then let it bloom.” The useful visual is the initial pour and the change in the coffee bed, rather than a later shot of the finished cup:

In “Pour-over Tutorial,” find the clearest overhead take of the initial wetting pour and the grounds bubbling during the bloom. Place the relevant portion on a video track above the presenter, starting as I say “wet all the grounds.” Keep my dialogue underneath. Show the bubbling while I explain the bloom, then return to the presenter for the next instruction. Use only uploaded footage for this passage. Leave the base cut, titles and music unchanged.

In Chris Ajtony's demonstration at 09:26 the screen recording on V2 covers the presenter footage on V1, while the presenter’s dialogue continues on A1.

This combines search and placement in one useful task. Searching footage by what happens in the shot is different from sourcing new material: here, ChatGPT is finding an action you actually recorded.

Review the chosen source range as well as its position. Does it show the beginning of the action? Can you see the detail the presenter names? For this example, a beautiful shot of the kettle is less useful than a clear view of the grounds changing.

If the match is wrong, narrow the search with a visible distinction:

Replace that selection with the overhead take where the whole coffee bed is visible and bubbles appear after the first pour. Exclude the later circular-pouring takes. Keep the overlay at the same explanation and preserve the dialogue.

Source stock footage for a missing visual

ChatGPT can also source stock supporting footage through the connected Editor. Tell it the subject, treatment and intended passage, and distinguish the request from searching your own files.

In this tutorial, the opening mentions starting with whole beans. If you have no suitable recorded shot, a stock close-up can establish that idea:

Source a stock close-up of roasted whole coffee beans in soft, natural light for the opening line about starting with whole beans. Choose a plain composition without visible packaging or text. Place it over that line and return to the presenter before the grind-size explanation. Keep the voice and music unchanged. Do not generate a new clip for this request.

The illustration shows a stock beans close-up covering the opening line, while the presenter’s dialogue and music continue underneath.

Stock works here because the shot establishes a general ingredient. Use the real recordings for the tutorial's pouring technique and results. If the project already contains a good beans close-up, ask ChatGPT to use that instead.

Watch the stock shot between its neighbouring clips. Similar light, framing and movement can help it fit. If it feels like a different video, ask for a more suitable selection or adjust its framing and colour on the timeline.

Generate an illustration for a difficult-to-show idea

ChatGPT can request new images or video for the project through invideo Editor, then place them in the edit. Describe the idea, appearance and role of the visual. Specify generation when you want a new asset, since finding an existing shot is a different request.

For the grind-size passage, a simple illustration can make a comparison easier to see. We will keep the actual grinding footage for the demonstration, and use an illustrative insert for the concept:

Generate a clean 16:9 illustration comparing a small pile of fine coffee particles with a small pile of coarse coffee particles, viewed at the same scale on a warm cream background. Use restrained brown and blue colours, generous space and no generated text. Add separate editable labels reading “Fine” and “Coarse.” Place the illustration above the presenter during the line comparing fine and coarse grind sizes. Keep it visibly illustrative, preserve the dialogue timing and return to my real grinding close-up for the demonstration.

The illustration shows the generated comparison on V2, with “Fine” and “Coarse” on separate text tracks above it. The presenter footage and dialogue remain underneath.

For a real Editor example, watch Matt Penny's demonstration at 05:03: generated supporting images sit on V2, below captions on V3 and above the base cut on V1/A1.

Generation uses invideo credits. Start with the visual the passage actually needs, rather than generating a full set of inserts before seeing how one works in the edit. Check that the comparison is clear and that any factual detail matches your explanation.

The generated image remains an image clip on the timeline. Its internal particles are not independently editable just because the clip is editable. Keeping “Fine” and “Coarse” as separate text layers lets you change those labels without regenerating the artwork. Our mixed-media editing guide explains that distinction in more detail.

If you need motion, ask for the specific movement, such as a slow move across the illustration, using editable keyframes. A still held long enough to understand may be all this passage needs.

Time each shot to the explanation

ChatGPT can adjust the overlays' start, end, source range and framing on the existing timeline. Give it the words or action that determine those boundaries. A relevant shot arriving before the subject is introduced can still confuse the viewer.

For the circular-pouring passage, ask for both the action and the timing:

Find the close-up that clearly shows the kettle moving in small circles. Place the close-up on a video track above the presenter footage, starting when I say “small circles from the centre outward.” Use a range that shows one complete movement. End it after the movement is clear, before I introduce the next step. Keep the presenter dialogue continuous, mute the close-up's source audio and preserve the existing music and captions. Frame it so the kettle spout and coffee bed remain visible.

The illustration shows the pouring close-up on V2 above the presenter on V1, while dialogue on A1 and music on A2 continue underneath.

The upper visual track changes what viewers see while the underlying presenter footage and dialogue stay in place. If you want the recorded pouring sound too, request that explicitly and balance it below the voice. Adding a shot does not require adding all its audio.

Play a few seconds before and after each insert. Check the subject's visibility, the time needed to understand the action, and the return to the main shot. Keep existing captions readable over the new picture. The aim is to explain the passage, not fill every second with another shot.

Refine a specific B-roll passage

Suppose the pouring insert starts on the right sentence, but the selected range misses the beginning of the circular movement. ChatGPT can change that source selection and its exit while preserving the approved edit.

Use current timeline timestamps, along with the shot and the reason for the change:

The illustration shows a request to adjust the pouring shot’s source range and duration while preserving the dialogue and approved layers.

These timings belong to the example. Use your own cut's boundaries, and check that the requested action fits before the next instruction. If it does not, choose a more suitable take rather than speeding up the demonstration until it becomes hard to follow.

You can also choose the exact in-point, trim the overlay or adjust its position manually on the frame-accurate timeline. The full editing suite remains available for your creative choices. ChatGPT reads the updated timeline before continuing, so you can build from those changes.

Review and export the finished edit

Play the full tutorial to check that every insert enters on the right explanation, lasts long enough to understand and returns cleanly to the presenter. Then identify the finished timeline for export.

Export the current “Pour-over Tutorial” timeline as a finished H.264 MP4 video. Keep the editable timeline and all its layers available for future revisions.

The illustration shows the finished-video MP4 option. Editable-project export is a separate choice when you want to continue in another editor.

Open the rendered MP4 before sharing it. Check the new visuals, captions and sound at the edited passages.

For an editable handoff, name the destination application when requesting project export. This keeps that request separate from the finished MP4.

This workflow is useful when your main cut already has a clear story and you want help carrying out the visual pass. You decide what viewers need to understand; ChatGPT helps find or create the material and fit it into the edit you are developing.

Frequently asked questions

Can ChatGPT add B-roll directly to my video?

For the editable-timeline workflow in this guide, ChatGPT needs a compatible video editor connected through MCP. In invideo Editor, it can search footage, source or generate supporting visuals, and place them on the timeline. A conversation alone can help plan the visual pass but does not reliably execute these edits.

Do I need to find and label every B-roll clip first?

No. Add the source footage to the connected project and describe the action or detail you need. ChatGPT can search by visual action or spoken content. Clear filenames can help, but you can ask for a moment without knowing which file contains it.

Is adding B-roll through invideo MCP free?

Ordinary footage analysis and timeline editing do not consume invideo credits. Your ChatGPT plan and usage limits apply separately. Generating new images, video or audio uses invideo credits.

Will B-roll replace my presenter audio?

It does not need to. Ask ChatGPT to put the supporting visual above the presenter footage, keep the dialogue underneath and mute the insert's source audio unless you want to use it. Check the sound through the insert's entrance and exit.

Can I change the shots manually after ChatGPT adds them?

Yes. The footage, image clips and separate graphics remain on an editable timeline in invideo Editor. You can trim, replace, reframe or retime them yourself, then continue with ChatGPT from the updated edit. Details baked into a generated image require revising that image or building separate elements.