One photo, one sentence, one fully packaged talking-head video

What it does

Digital Human Koubo is an Agent Skill (a SKILL.md plus Python/shell/Remotion scripts) that automates the whole pipeline from a still photo to a finished talking-head video: script, lip-synced presenter clips via Grok Video 1.5 through the Renoise MCP, voice clean-up, captions, and YouTube-style packaging. The presenter is a persistent persona stored in persona.json, so every video features the same face, outfit, and voice. Final assembly happens in Remotion 4 with one of five editing styles.

The interesting bit

The boring middle is where the engineering lives: every generated segment is transcribed with whisper and compared against the script, plus 4 frames per segment are inspected because Grok sometimes burns subtitles into the picture — bad segments get re-shot. Thirty separately generated clips are made to sound like one recording via spectral matching, de-essing, and a shared room reverb (sibilance spread reportedly 13.2 dB → 5.9 dB). And the agent quotes segments · seconds · credits · balance before spending anything, waiting for your “yes”.

Key highlights

• 48 packaging components measured frame-by-frame from 25 videos by top creators (MrBeast, MKBHD, Johnny Harris, 何同学, and more)
• 5 editing styles, with the agent diagnosing your content and recommending two before asking everything in one go
• Text-behind-presenter compositing via macOS Vision person segmentation into a 3-layer stack
• Multi-language output — EN/JA/KO/FR/ES/AR tested, with whisper back-checks at 94–100%
• Works with any agent that can load skills, connect a remote MCP server, and run local commands (Claude Code tested; Codex, OpenClaw, WorkBuddy expected)

Caveats

• Depends on paid Renoise credits (the demo ad cost 482) and a browser login — no API-key path
• The text-behind composite relies on macOS Vision, so that feature is platform-bound
• Several agent integrations are marked “should work” rather than tested

Verdict

Worth a look if you produce talking-head content and already live in an agent workflow — the QA and cost-gating are genuinely thoughtful. If you wanted a self-contained, offline pipeline, this is orchestration over external paid services, not that.