Claude Opus 5.5 · AI Video Editing with Borumi

Quick Summary

• Paul J. Lipsky records camera, microphone and screen inside Borumi, divides the tutorial into scenes, and lets the app build a synchronized timeline with a searchable transcript.
• Claude Code reaches that project through Borumi's local MCP integration. Opus 5.5 can inspect the timeline and transcript, remove silence and false starts, switch layouts, crop exposed interface elements, add zooms and use screen highlights.
• The reusable skill is intentionally simple: repeat the same cleanup rules for every project, let the model make a first pass, then review the result and request targeted corrections.
• Motion graphics are directed separately. Lipsky talks through the timeline, describes the composition and the purpose of each graphic, and asks Claude to insert the result at the right moment.
• The most surprising demonstration is a missing screen recording. Claude captures the requested product page and places that footage into the relevant scene instead of forcing a new manual recording session.
• Lipsky says the automated pass gets him about 80% of the way to a final edit. He still checks every cut, removes unwanted zooms, corrects mistakes and makes subjective visual decisions himself.
• Borumi's official documentation confirms the Script, Record and Edit workflow, transcript editing and MCP support. The creator discloses that his link is an affiliate link, although the video itself is not sponsored.

Credit and the real meaning of 'best editor'

This workflow comes from Paul J. Lipsky's embedded video, published on September 28, 2026. He demonstrates the complete process inside Claude Desktop and Borumi, and says the example video and the YouTube tutorial itself were edited with Claude Opus 5.5.

The title is a personal verdict, not a universal benchmark. Lipsky means Opus 5.5 gets closer to his preferred finished result than the other models he has tried. His clearest measurement appears near the end: the automated pass gets him about 80% of the way there, after which he performs a final human polish.

That qualification is the useful lesson. The model can execute a large amount of technical editing, but the creator still decides what feels slow, which zoom is distracting, how a graphic should look and when a scene has enough visual movement.

The stack: Claude Desktop, Claude Code, Opus 5.5 and Borumi

Lipsky works in the Claude desktop app and normally uses Cowork for general tasks. For video editing, he switches to Claude Code because it gives the model more direct control over the connected project. He selects Opus 5.5 with high thinking effort.

Anthropic describes Opus 5.5 as its strongest Opus model for coding, agents and complex professional work. That agentic capability matters here because the task is not one generation. Claude must read project context, plan a series of edits, call Borumi's tools, inspect the result and respond to revisions.

The creator says his $100 monthly Claude plan covers his own weekly editing and other work. That is an individual usage report, not a predictable cost for every creator. Plan limits, model access and usage can change, so check the current Claude account details before budgeting a production schedule.

Why Borumi is a good fit for this specific job

Borumi combines planning, recording and editing in one desktop application for macOS and Windows. A structured project has three main areas: Script, Record and Edit. The script is optional, but it can divide a tutorial into named scenes and supply a teleprompter.

In the Record view, camera, microphone and screen are separate sources. The screen capture can target the full display or one application, and system audio can be recorded when a browser demo or video needs sound. Because every scene is recorded independently, a failed explanation can be retaken without repeating the whole presentation.

Once recording is complete, Borumi creates the timeline and transcript. Deleting words from the transcript removes the matching video. The editor also supports layouts, cursor-following zooms, screen highlights and separate framing for face and screen. Those are precisely the controls Claude later operates through MCP.

Understand the manual edit before automating it

Lipsky first shows the edit he used to perform by hand. He deletes dots that mark long pauses, reads the transcript for false starts, keeps the cleaner repeated take, switches between screen-only and presenter layouts, then adds zooms where the viewer needs to see a click or model selector.

He also uses a highlight effect to darken the rest of the screen around one important region. This manual tour is more than product orientation. It defines what the AI should be allowed to change and gives the creator a vocabulary for reviewing the result.

A reliable automation starts with an explicit baseline: what counts as dead air, which repeated take should survive, how much natural breathing room to preserve, which interface chrome may be cropped and when a zoom adds clarity. Without those rules, 'edit this video' leaves too much room for arbitrary choices.

Connect Borumi to Claude through MCP

Borumi added its MCP integration in version 0.29.0 on July 31, 2026. Its official changelog says the connection lets Claude, ChatGPT or another compatible AI help script, edit and export projects. The integration requires a supported desktop client and does not work through the ordinary web versions because the Borumi server runs locally.

Inside Borumi, open Settings and choose AI, then follow the setup instructions for Claude. Restart the AI client if requested, open the Borumi project and tell Claude the exact project name. The video uses a project called AI News Demo.

Version history is important when an agent can modify a real timeline. Borumi introduced project versioning with the MCP release, so an unwanted pass can be rolled back. Even with that safety net, save the project and inspect staged changes before exporting.

Turn repeated cleanup rules into one small skill

The first pass is deliberately unglamorous. Lipsky asks Claude to remove long silences and bad takes, clean the outer edges of the screen capture, hide exposed menu bars and add zooms where he clicks or points. Because those instructions repeat on every project, he stores them in a skill called Edit Borumi Video.

The skill is not presented as a giant secret prompt. Its value comes from consistency. The same baseline rules are applied to every project, and project-specific layout or emphasis notes are added afterward. Lipsky often lets the model finish a broad first pass, watches the video and then requests precise changes.

A useful brief should also state what not to change. Preserve natural pauses below a chosen threshold, do not cut between linked clauses, never cover important interface text with the presenter bubble, and do not add zooms merely to create activity. The prompt pack below turns the demonstrated method into a reusable editorial template.

Direct motion graphics by purpose and beat

After cleanup, Lipsky scrubs the timeline and dictates the creative brief. For the opening scene he asks for a split composition with the presenter on the right, a weekly AI news title on the left and proof from his YouTube channel. For the usage comparison he requests a data graphic that follows the spoken numbers.

The instruction contains four useful layers: the scene, the communication goal, the layout and the evidence. 'Add motion graphics' is vague. 'Keep me on the right, introduce the weekly recap on the left, then reveal my channel as proof that I cover this every week' gives the agent a sequence it can design against the narration.

Lipsky says his older workflow used one model for the timeline and Claude for graphics, followed by manual placement. In this demonstration, Opus 5.5 creates and inserts the graphics in the same pass. The result is still reviewed as an edit, not accepted because the agent says it is complete.

Claude can fill a missing screen-recording beat

One scene discusses a new iPhone, but Lipsky did not capture supporting footage. He asks Claude to open the Apple page, record the product images while clicking through them and insert that recording while he speaks. The finished timeline includes the requested visual coverage.

This is powerful because it closes a gap without reopening the recording session. It also needs more supervision than silence removal. The page may change, the capture can include private tabs or notifications, product details may be misidentified, and third-party media can have usage restrictions. Use a clean browser profile, give the exact approved URL, specify the crop and duration, and review every captured frame before publishing.

The review loop is the workflow, not an afterthought

When Claude finishes, Lipsky skips the text report and opens Borumi. He checks whether silences and false starts are gone, confirms the final take survived, watches layout changes, tests zoom timing, inspects screen highlights and plays every generated motion graphic in context.

The agent pass is about speed. The final pass is about judgment. Lipsky removes zooms he no longer wants, adjusts sections that feel visually wrong and corrects small mistakes. That last 20% contains much of the perceived quality because it controls rhythm, restraint and whether an effect helps the viewer understand the tutorial.

Before export, watch once for story and pacing, once for visual errors and once with headphones for audio. Check that crops do not hide menus being discussed, presenter overlays do not cover important controls, screen recordings show the intended page, graphics match the spoken numbers and every transition leaves enough time to read.

Pricing, trial and disclosure

Lipsky states that the video is not sponsored and clearly discloses that his description contains an affiliate link. The official Borumi site currently lists a free trial and an early-adopter lifetime license at $79 instead of $99, with a 30-day money-back guarantee. It says the trial lets you use the application before paying to export.

Those prices and terms were checked on September 28, 2026 and can change. The Borumi license does not include the separate cost of Claude. The official site also notes that future cloud features may offer basic access to lifetime customers with optional paid add-ons.

The cleanest evaluation is to test one real project. Record a short tutorial, run the same cleanup brief you would give a human editor, time the review and correction pass, then compare the saved labor with the combined software cost. A fast first pass is only valuable if fixing it takes less time than editing manually.