AI in DaVinci Resolve: What It Edits Well and Where It Fails

Artificial intelligence can now edit video inside DaVinci Resolve, one of the most widely used editing programs in the world: it cuts, assembles, adds titles and exports from a request written in plain language. In a public test, it turned 24 minutes of raw footage into a 7-minute cut in about 15 minutes. But it still makes mistakes that viewers notice, and that determines who this new feature is worth it for.

What changed in DaVinci Resolve 21.1

DaVinci Resolve version 21.1 was released in September 2026. The big news for anyone who is not a professional editor is a connector built into the program, based on MCP, an open standard that lets an AI assistant trigger functions inside another piece of software instead of just talking about them. In practice, the user writes the request in the assistant's chat and the AI carries out the work on the editor's timeline.

Three details matter before any excitement:

• It works only in the paid version. The connector exists only in DaVinci Resolve Studio, sold for US$ 295 as a one-time payment. The update is free for anyone who already has the license.
• The free version was left out. The free edition is still very capable for editing, but it does not have the connector and lost Python automation, which moved to Studio. For that reason, tutorials that teach how to connect AI to the free version through third-party programs may no longer work. Test before you count on it.
• The AI is purchased separately. Resolve works with assistants such as Claude, Claude Code and ChatGPT's Codex, and each one has its own subscription or usage-based billing.

The manual from Blackmagic, the program's maker, includes examples of what to ask for: generating files in several formats, renaming clips in sequence, marking segments and exporting the list, applying a standard color to all clips, checking delivery requirements and reframing the timeline to the vertical 9:16 format used by Reels. That last request alone is already of interest to many people who shoot horizontally and publish vertically.

The test that gives the order of magnitude

The most useful report so far comes from Justin Brown of the Primal Video channel, who published a step-by-step account with timings. It is a test by one person, with one type of video, and not a general measurement. Even so, it gives a good idea of what to expect:

• Initial cut: 23 minutes and 58 seconds of footage became about 7 minutes and 14 seconds, in roughly 15.5 minutes of AI work. It picked the last take of each segment, made hard cuts and zooms, and left markers for review.
• Animated titles: they came in synced with the speech, with key points pulled from the transcript.
• Visual identity: after receiving a reference page, the AI redid the graphics in that style in about 8 minutes.
• Export: it came out in 4K, and the AI itself checked resolution, frames per second and duration.

The estimated cost of the run was between US$ 13 and US$ 19 in AI usage, plus a few cents for transcription. This is the author's estimate, and the amount varies with the model and the plan.

Where the AI still makes mistakes

The same test shows the flaws, and they are the kind the audience notices. Repeated segments and wrong sentence starts were left in, some cuts kept breaths and noises before the speech, and the titles, without brand rules, came out correct but generic.

The main cause was the transcript. Resolve's own transcript skipped parts of the speech, and the AI cuts guided by the text: if the text is wrong, the cut comes out wrong. When the author supplied a better transcript, with the timing of each word, the AI redid the assembly in under 5 minutes and the result improved with no other instruction. His conclusion is straightforward: the quality of the transcript determines the result.

There are also technical limits. PilotCut, a competing company and therefore an interested source, points out that Resolve's automation does not yet offer fine control over adjustments such as trimming a clip frame by frame or changing the speed of a segment, and that some commands seem to work but get lost on save. Treat it as a warning to test, not as a verdict.

And it is not instantaneous on a demanding project. A report about the workflow of the editor known as Future Brian describes 14 rounds of revision over a day and a half for a single video, with about 25 minutes of rendering in each round.

What you can hand to the AI and what stays with people

The split that appears in the more serious reports is similar. The AI performs well on manual, repetitive work:

• organizing files and renaming clips;
• assembling the first cut of a talking video;
• removing long pauses and wrong takes;
• generating versions in other formats, such as vertical;
• repeating exports at each round of adjustments.

The decisions that give the video meaning remain with a person:

• what the video needs to say, to whom and in what tone;
• which segment is moving and which is surplus;
• the color finish and the final pacing;
• the review before publishing, watching the whole video.

In other words, the AI replaces hours of cutting, not the judgment of whoever decides what stays.

Is “old-school” editing over?

The question has become a public fight. Future Brian posted a video on X with that provocation using AI inside DaVinci, which drew 1.2 million views, and later said he received 5 death threats and 44 blocks. The reaction shows how big the fear is among professionals.

The honest answer has two parts. For talking videos, manually cutting pauses and mistakes does tend to disappear: it is the most tedious part and the easiest to automate. For an advertising film, a documentary or any piece where the pacing is the message, the editor still decides, and now has more time for what matters. Whoever knows how to edit is precisely the person who notices when the AI got it wrong.

Is it worth it for a small business?

The math depends on volume and on the type of video.

It tends to be worth it if you record often, always in the same format, speaking to the camera: tips, lessons, testimonials, product explanations. In that case, US$ 295 once, plus the AI, can pay for themselves in a few weeks of saved hours.

It tends not to be worth it if you publish a video once in a while, depend on scenes with mood and movement, or do not have a computer capable of running the program. Resolve does the editing and exporting on your machine, and the learning curve exists, even with the AI helping.

There is also a third path. For videos with a fixed format, such as animated tips, generating the video with code can be simpler than editing, as we explained in video edited with code. For a film with people, product and setting, the limit is usually in image generation, the subject of advertising video with AI.

Precautions before letting the AI touch the project

• Work on a copy. Duplicate the project before each new type of request, until you trust the result.
• Protect the original files. Blackmagic itself warns that scripts created by the AI have the same access to the disk and the network as the assistant. Keep a copy of the footage somewhere the AI cannot reach.
• Be careful when updating. There are reports that projects opened in 21.1 no longer open in the previous version. Make a copy first.
• Check what you are sending. With client footage or material under contract, read the AI service's terms and what was agreed with the client before connecting.
• Watch the whole video. No public measurement shows how often the AI gets it right. The final review is still yours.

How to test it in five steps

1. Start with an unimportant video. Use an old talking recording, 5 to 10 minutes long, that is not going to air.

2. Record with editing in mind. Clean audio and a steady structure help more than any command. Redid a sentence? Repeat the whole sentence, so the last take is complete.

3. Take care of the transcript. If the cut comes out with errors, swap in a better transcript before changing anything else.

4. Write the request as if you were talking to an editor. State the audience, the length, the tone and what must not be left out. Show an example of the style you want, such as a page from your website.

5. Compare with the time you used to spend. Note how long it took, including the review. If the math does not work out, the AI is not yet for your case.

If your problem is having good videos every week without becoming an editor, the alternative is to leave it to people who do it every day. See how we work on the promotional video production page and find out why short video sells more.