Automating DaVinci Resolve with GPT-6 Astra: 14-Year Editor's Full Workflow
A breakdown of 14-year video editor @EHuanglu's workflow using GPT-6 Astra Computer Use to automate DaVinci Resolve: cutting 84 clips, library BGM matching, and
On September 11, 2026, veteran video editor @EHuanglu, bringing 14 years of professional cutting experience to bear, shared an end-to-end video editing automation workflow on X powered by GPT-6 Astra's native Computer Use capabilities. Operating without custom plugins or complex scripting layers, the artificial intelligence model navigates the desktop interface of DaVinci Resolve directly—ingesting 84 raw video clips, trimming timeline footage, selecting local library background music, and generating synchronized subtitle tracks entirely on its own.
Image source: @EHuanglu (X)
Video post-production has long been regarded as exceptionally challenging for artificial intelligence to master. While generative models have proven capable of synthesizing isolated shots or summarizing scripts, assembling long-form sequences in a Non-Linear Editor (NLE) demands continuous temporal pacing, audio synchronization, and visual context across multiple video and audio tracks. However, this production workflow demonstrates that a vision-enabled desktop agent can directly execute the full mechanical burden of video editing, taking uncurated footage from raw intake to a polished, multi-track timeline assembly.
Automated Assembly of 84 Raw Clips and DaVinci Resolve Timeline Cutting
The core technical distinction of @EHuanglu's workflow lies in bypassing rigid Model Context Protocol (MCP) wrappers and brittle API layers in favor of native visual Computer Use. The model views the operator's display and executes mouse movements and keyboard shortcuts just as a human editor would.
- Direct desktop app control: Astra launches DaVinci Resolve on the workstation, visually inspects the media pool, and controls the edit timeline directly. When asked in community replies about system requirements, @EHuanglu confirmed that Astra interacts with standard desktop applications installed locally without specialized server infrastructure.
- Curating and trimming 84 raw clips: Rather than requiring human pre-screening, the AI agent watched all 84 raw video clips sequentially, excised shaky takes and unwanted footage, and trimmed the highest-impact action into coherent chronological scenes on the primary timeline.
- Autonomous background music matching: Instead of relying on cloud audio services, Astra browsed the editor's local sound library, evaluated available tracks against the sequence's tempo, and aligned the chosen music track beneath the video cuts.
- In-timeline subtitle generation: Astra parsed spoken dialogue directly within DaVinci Resolve, creating dedicated subtitle tracks and locking text placement precisely to corresponding speech timecodes.
Having cut video professionally for 14 years, @EHuanglu noted that watching an autonomous agent complete multi-track timeline editing was sobering, demonstrating that manual cutting tasks can now be delegated entirely to AI.
End-to-End Production Pipeline from Canvas Assets to Timeline Assembly
Beyond slicing and arranging pre-existing camera footage, the demonstration outlined a broader end-to-end production architecture spanning creative concept generation through finished video delivery.
Astra directly navigated a web-based creative canvas interface to generate project character sheets and individual scene video clips with visual style consistency. These generated assets were then imported directly into DaVinci Resolve's media pool, establishing an unbroken link between generative asset creation and post-production editing.
- Unifying visual generation and NLE assembly: Rather than forcing a human operator to download, rename, and manually import generated visual assets between isolated web tools and editing software, the autonomous agent bridged both environments directly.
- Public access to prompts and logs: To substantiate the workflow and enable community replication, @EHuanglu released the complete conversational chat logs and execution prompts for both canvas asset creation and DaVinci Resolve desktop automation.
Reproducing Editing Taste and Evolving Studio Production Dynamics
Addressing the longstanding critique that artificial intelligence lacks intuitive artistic taste and rhythmic editing sensibility, the workflow highlighted practical techniques for replicating human editorial pacing.
- Replicating pacing through style references: While critics frequently argue that AI systems lack genuine taste, @EHuanglu demonstrated that editing rhythm and pacing can be effectively replicated by supplying the model with a concise set of well-cut reference clips to serve as visual style guides.
- Restructuring commercial production workflows: Commercial brands and small creative agencies that previously relied on costly external post-production houses can now empower a single internal team member, such as an intern, to oversee autonomous editing agents that produce finished commercial assets.
- Production guardrails for autonomous operation: To ensure long-running autonomous sessions remain stable and avoid timeline discrepancies or unintended track collisions, editors should establish explicit pacing rules and audio threshold boundaries within their initial prompt constraints.
Original source
Primary workflow demonstrations, execution logs, and full prompting materials provided by the author:
- @EHuanglu on X: 2026-09-11 DaVinci Resolve Automation with GPT-6 Astra Thread
- @EHuanglu Chat Log: GPT-6 Astra Execution Log and Session Details
- @EHuanglu Prompts and Workflow Guide: Canvas Asset Generation and DaVinci Resolve Editing Prompts