Miadi Review

Engineering an Autonomous Video Editor with Claude Code

Oct 10, 2026 · model: manual-edit · channel: Christian Peverelli

Watch source

Engineering an Autonomous Video Editor with Claude Code

Source: I Fully Automated My Video Editing Using Claude Code (Full Walkthrough) — Christian Peverelli https://youtu.be/HzXD4GVqXwM

Summary

Christian Peverelli demonstrates how to construct an automated video production system orchestrating Claude Code across local CLI tools and external APIs. By decoupling video editing into specialized functional nodes—Parakeet for transcription, HeyGen HyperFrames for HTML/CSS-rendered 4K motion graphics, Tella via MCP for screen capture, and FFmpeg for final assembly—Claude acts as an executive project manager. The workflow codifies editorial standards into a persistent SKILL.md file, using timestamp-directed critique to continually refine instructions into deterministic rules. This architecture eliminates manual timeline trimming, slashing turnaround times from days to hours while reducing cost per video to approximately $27.

Intended Use & Continuation

Study each steps he goes thru and ours into the whole process of screenwalk, miadi-chronicle, etc and get inspired on steps we can enhance and use providers that he talks about that we might need. {{ edit this file and include these steps summary in subsections bellow and their relation to other components we already have and how they integrate max 55 words for step and 55 words for relation }}

—-

Concepts

  1. Automated Video Editing: The core idea of using Al to replace a traditional post-production team (0:00-1:06).
  2. The "Skill" File: A foundational instruction set, consisting of approximately 1,400 lines, that defines editing preferences, brand styles, and rules for the Al (5:33-6:07).
  3. Human-in-the-Loop: The creator's role as director, providing feedback on Al-generated drafts to refine the system and improve future outputs (10:35-11:13).
  4. Cost Efficiency: Transitioning from expensive treelance editors (250-1,000+ per video) to an automated system costing roughly $27 per project (15:12-16:07).

Capabilities

  1. Local System Management: Claude Code operates locally on the user's computer, managing files and coordinating various software tools (1:07-1:21).
  2. Intelligent Transcription: Automated transformation of raw footage into text with accurate timestamps for easier cutting (1:34-2:12).
  3. Dynamic B-roll Integration: Researching and grabbing web screenshots or video clips to visually support claims (2:47-3:44).
  4. Automated Finalization: Using FFmpeg to handle technical quality control, such as removing black frames and checking audio sync (5:04-5:28).
  5. Accelerated Production: Reducing total workflow time from 3-7 days to same-day planning, shooting, and releasing (16:11-16:34). generating transcriptions and timestamps (1:36-2:00).
  6. Hyperframes (Animator): A tool to transform Claude-written code into sharp 4K motion graphics (2:13-2:46).
  7. Tella (Screen Editor): Records screen-shares with an MCP connector for automated layout and formatting (3:44-4:19).
  8. Epidemic Sound (Sound Designer): A searchable, integrable database trained to apply preferred music and sound effects (4:20-5:04).

Where can I find the skill mentioned ?

You can find the editor skill file mentioned in the video within the creator's free community called We Are No Code. The creator provides the link to this community in the video description (9:35 - 9:40, 16:36 - 16:43).

—-

Aspects

Local Orchestration via Claude Code (00:59-01:26)

Claude Code operates directly on the user's local machine, serving as an executive project manager. It reads source files, invokes system utilities, executes terminal commands, and routes output assets into structured directory targets without cloud file uploads or manual software programming.

Decoupled Multi-Tool Agentic Graph (01:34-05:28)

The architecture delegates discrete editorial jobs to specialized modular utilities: Nvidia Parakeet for timestamped speech-to-text transcription, HeyGen HyperFrames for code-based 4K motion graphics, Chrome for online b-roll capture, Tella via MCP for interactive screen recordings, and FFmpeg for headless media assembly and sanity checks.

Declarative Brand Rule Codification (05:31-06:03)

Editorial taste, typography, color palettes, and motion guidelines are documented in a centralized SKILL.md instruction file. Expanding to 1,400 lines and dozens of explicit rules, this file translates subjective creative intuition into declarative engineering guardrails that Claude references before executing edits.

Feedback Ingestion and Skill Hardening (09:43-10:47)

The creator reviews draft cuts using timestamped natural language notes. Instead of remaining one-off corrections, accepted fixes are programmatically written back into the skill definition file, creating an iterative reinforcement loop that hardens rules and prevents the system from repeating past mistakes.

End-results and Core Aspect

Core Aspect — Autonomous Overnight Video Compilation Pipeline (14:36-15:11)

The system achieves fully autonomous video creation where unedited camera takes are fed directly to Claude Code. While the creator sleeps, the agent isolates optimal takes, compiles motion graphics, synchronizes audio tracks, checks visual parity, and outputs publication-ready 4K files without manual timeline interaction.

Results

  • Same-Day Production Turnaround (16:11-16:24): Compresses traditional editing cycles from 3 to 7 days of asynchronous human communication down to same-day delivery, allowing creators to record, automatically process, review, and publish content within hours.
  • Over 90% Cost Reduction (15:30-16:10): Reduces production expenses from standard freelance rates of $250 to $1,000+ per cut down to approximately $27 per finished video, comprising $24 in model token inference, $2 in screen recorder fees, and $1 in music licensing.
  • Zero-Touch Multimodal Render Execution (13:19-13:49): In demonstration, the pipeline converted 2 minutes and 46 seconds of raw, mistake-filled footage into a calibrated 31-second 4K master in 16 minutes, seamlessly placing graphics, audio swells, and zoom framing to spec.