Can AI Actually Edit a Video in DaVinci Resolve? It Already Has.

Brent Schooley, who works on developer experience at OpenAI, posted this week that OpenAI's own launch video for GPT-Image-2.5 Flare and Sunburst was edited by an AI agent inside DaVinci Resolve. Not generated. Not assembled from templates. Edited, in the same application working editors already use, with him reviewing and directing the process rather than executing it by hand.

The Actual Workflow

Schooley described the process directly in response to a follower's question: he records with a teleprompter script, then hands that script along with all camera footage, audio, and screen recordings to a new Codex thread. He asks it to edit the video in DaVinci Resolve. After the first cut comes back, he gives notes and repeats the cycle.

That's a notably different workflow than a single "generate my video" prompt. It's iterative, script-first, and keeps a human reviewing and redirecting at every pass, closer to notes on a rough cut than a one-shot generation request.

Why This Specific Video Matters

This wasn't Schooley's first attempt. Back in July, he used the same general approach, Codex driving an edit in DaVinci Resolve, for a video celebrating the OpenAI YouTube channel crossing 2 million subscribers. In that earlier project, he noted the AI inspected video frames directly to verify launch clips stayed in the correct order, product labels matched the right footage, and no bad transitions or single-frame flashes slipped through, then picked the background music itself, explaining its reasoning: a track selected specifically for "a spacious opening, a clear pulse for the launch montage, and an uplifting finish."

The GPT-Image-2.5 video is a continuation of that same practice, now applied to a real, current product launch rather than an internal milestone video.

What DaVinci Resolve 21.1 Actually Changes for This

Schooley published a detailed technical breakdown of what Resolve's new scripting APIs unlock specifically for agentic editing, and it maps directly onto the workflow he just used.

The new MediaPoolItem.GetTranscription() API exposes timed words, segments, and speaker labels directly from the project, letting an agent connect a request like "make a 60-second explainer about this product" to specific sections of source material without needing external transcription first. For dialogue-heavy editing, interviews, podcasts, explainers, Schooley calls this "huge."

New multicam scripting lets an agent create multicam clips, align footage, run SmartSwitch, and flatten the result, with controls for minimum shot duration, switching delay, and wide-shot frequency exposed directly. That means a prompt like "keep the cuts relaxed and use occasional wide shots" becomes something an agent can actually execute with precision rather than guessing at pacing on its own.

Expanded audio APIs cover normalization, channel mapping, and controls for voice isolation and dialogue leveling, letting an agent both fix inconsistent levels and verify the fix actually applied correctly. Transitions, fades, and clip-speed controls also became directly scriptable. Previously, Schooley notes, these specific tasks often required unreliable Computer Use, an agent clicking through the interface rather than calling a defined function, which made "add short fades, slow this B-roll clip" a much less reliable request to fulfill.

The Timeline Detail Worth Knowing

Schooley was direct about one detail that matters for anyone trying to replicate this: the GPT-Image-2.5 video was built using his own custom CLI, before Resolve's new native MCP server existed. Asked whether he used the new built-in Studio 21.1 capabilities or a third-party GitHub MCP, he clarified plainly: "This one was before the new MCP with my own CLI." Going forward, though, his stated plan is clear: "I'll happily delete mine... team Resolve MCP."

That's a meaningful signal on its own. A developer experience professional who already built and maintained his own custom tooling for this exact purpose is choosing to abandon it in favor of the native solution, specifically because it will stay current with each Resolve release rather than requiring ongoing maintenance against a moving target.

Competitive Context

This is a concrete answer to a question BRC's coverage has been circling for months: whether AI agents can actually perform the fine-grained timeline work of editing, not just generate raw footage or a rough assembly.

Schooley's account, backed by a real, shipped OpenAI launch video and a detailed technical explanation of exactly which new APIs made it possible, is a genuinely stronger data point than a single self-reported demo, since it comes from someone with direct professional reason to actually use this in ongoing production work, not a one-off experiment.

The Signal in the Noise

The most important detail here isn't that AI touched a real timeline, it's how deliberately human-supervised the process still is. Script first, footage handed over explicitly, a first cut reviewed and redirected with notes, repeated until it's right. That's not autonomous filmmaking. It's a genuinely capable assistant editor that still answers to a director, which is a meaningfully different and more immediately useful claim than "AI can make your video for you."

Would a workflow like this, iterative, script-first, with you reviewing every cut, change how you'd think about bringing an AI agent into your own editing process?

Specs & Pricing

• Editing agent: OpenAI Codex, driving DaVinci Resolve via scripting
• Workflow: teleprompter script plus all footage/audio/screencasts provided upfront, iterative review and revision after each cut
• Key new Resolve 21.1 API: MediaPoolItem.GetTranscription(), exposes timed words, segments, and speaker labels
• Also new: scriptable multicam creation and SmartSwitch controls, expanded audio cleanup APIs, direct scripting for transitions/fades/clip speed (previously required unreliable Computer Use)
• Access note: native MCP server ships with Resolve Studio 21.1; this specific video was built on a custom CLI predating that release

Resources & Reads

• Brent Schooley's technical breakdown: DaVinci Resolve 21.1 Gives AI Editing Agents Better Tools
• BRC: DaVinci Resolve Can Now Take Direct Orders From Claude and ChatGPT
• BRC: GPT-6 Astra Turned One Prompt Into a Fully Edited, Publish-Ready Video