Claude and DaVinci Resolve: AI prepares the rough cut, the editor owns the result

October 23, 2026 update: OpenEdit and audio checks extend the editing workflow

Jirka Herník highlighted OpenEdit on October 22. That dates the recommendation, not the project’s launch. VEED’s official repository describes HTML composition rendered in a browser. Transcription needs word timestamps; choosing local WhisperX or hosted transcription changes where media is processed.

Alongside Resolve timeline preparation, this is a candidate for repeatable deliverables: captions, a short product demo or a training-video variant. My pilot would use one approved clip, fixed wording and two output sizes. Acceptance would require correct product names, readable captions and preserved meaning. Keep source media and project files with the export.

Mastermind 45 also points to Earshot, a local CLI, Python API and MCP toolset for describing, measuring and assessing audio. earshot measure reports measurements; earshot critique provides timestamped feedback. Its optional cloud path is separate.

I would test quiet speech under music, a clipped sentence ending and an effect landing away from the cut. The agent should identify the location and propose a correction; a person still listens to the final video. A score based on reference films is not a universal measure of training quality.

I have not benchmarked either tool here. Measure time to an accepted version, including corrections. n8n or Make could track the job and review handoff; this does not claim a ready-made integration.

---

Business video often becomes expensive before creative editing even starts. Someone has to organize footage, locate useful statements and prepare a coherent handoff. That is a good place to try AI assistance.

Filip Oborník’s AI s rozumem video, published on 19 October 2026, describes using Claude Code through MCP to prepare transcripts, multicam material and a script-based rough cut from roughly 70 clips in DaVinci Resolve. He reports about an hour of work. This is his case study, based here on public descriptions and chapters—not our benchmark.

Aim for an editable handoff

For a business pilot, I would ask for an editable project with traceable source material. A product interview, internal training session or customer story would work well. The agent proposes selections; an editor checks meaning, pacing and continuity.

The technical bridge is the community samuelgursky/davinci-resolve-mcp project. It exposes Resolve scripting tools and documents Studio, local external scripting and setup through npx davinci-resolve-mcp setup. It is not an official Blackmagic plugin.

Record the Resolve, client and MCP versions before testing. Use a duplicate project to verify that the agent can inspect media and create a disposable timeline. A successful installation does not establish editing quality.

Free editions need a version check

The project README describes a bridge for free Resolve 21.0.x and Python scripting restrictions from 21.1. Do not assume the video’s approach works in every current free installation. Check Blackmagic’s support and release information for your environment.

I would not start a team pilot by downgrading an active project. Establish compatibility and a working backup first. If a required operation is unavailable, the agent should report the limitation and stop.

A pilot from footage to review

The following is my suggested operating procedure, not a transcription of the video.

1. Package the input. A working copy of the footage, script, target duration, required messages and excluded shots. Give every clip a stable ID.
2. Build an index. Tie transcript passages to clips and time ranges. Flag uncertain words and check product names and numbers against the brief.
3. Propose the assembly. Produce a selection list with reasons before creating the working timeline. Preserve originals.
4. Check continuity. Look for truncated sentences, misleading B-roll and audio synchronization errors.
5. Hand it over. Include the project, preview, unresolved questions and a short action log.

A useful brief might say: “Prepare a three-minute working cut from this script. Link every selected statement to its source clip and time range. Flag uncertain transcription. Preserve source media and leave publication to the editor.”

Measure the accepted rough cut

Track time to an accepted rough cut, meaning-changing corrections, synchronization failures and missing required shots. Include setup, model calls, transcription and human review in the cost. A fast agent run followed by hours of repairs is not a saving.

If needed, n8n or Make could manage the surrounding job record: approved input links, assignment and a “ready for review” status after handoff. That is an integration proposal, not a claim that a native connector or tested automation is provided here.

Decide in advance which client recordings may reach a model or transcription service. Running the editor locally does not establish that every processing step stays local.

My starting point would be the preparation an editor repeats on every job. Keep story quality, faithful quotations and publication approval inspectable.

The cover is a photorealistic AI illustration.