lanshu: one authorized portrait and a script in, a QA-passed presenter video out

“Drop in a script, get a digital-human presenter video” — that is the promise, and lanshu-create-ai-presenter-video by Lanshu (MIT, 2,924 stars as of November 4, 2026) delivers it as an agent skill rather than a closed product: hand a topic or script to your coding agent and it runs the whole production line — script, narration, presenter generation, lip-sync, captions, keyword motion graphics, editing, rendering and quality assurance. Where one-click SaaS tools lock you in, this one is harness-neutral (Claude Code, Codex, Gemini CLI, Cursor and friends all load it) and provider-neutral (voice, presenter video, lip-sync and speech recognition are picked by capability at run time, with each job logging the vendor, model, parameters and task IDs it actually used).

Core features

• Two routes: the presenter route generates a digital human from one authorized portrait; the styled route needs no human likeness — narration is performed scene by scene in one of twelve styles (nine visual, three narrative).
• Evidence-gated stages: every stage must leave evidence on disk, and nothing is delivered until the output passes decode and loudness checks — “verified, publish-ready” is enforced, not claimed.
• Provider bookkeeping: every video records the providers, models, parameters and task IDs actually used, so cost and quality can be compared across runs.
• Render pipeline: rendered with HyperFrames; needs FFmpeg and Python 3.9+, plus Node.js for the styled route.
• Standard skill shape: a standard SKILL.md that any agent able to read files and run shell commands can follow directly.

Typical use cases

• Knowledge and news channels: one authorized portrait, daily scripts, a stable on-air style.
• Product explainers: pick a styled route and turn documentation into a narrated short film.
• Multi-version production: swap voice and subtitles for locale variants while the QA gates keep quality flat.

Quick start

git clone https://github.com/cclank/lanshu-create-ai-presenter-video.git

Hand the repo to your coding agent and name your route; install FFmpeg, Python and (for the styled route) Node.js, then drop in a script. Stage evidence stays on disk, and nothing ships until QA passes.

Summary

lanshu fits solo creators and small teams that need steady volumes of presenter or explainer video; broadcast-grade cinematography and no-terminal users are better served by closed SaaS. Facts and limits: MIT at the repo level, with Python/FFmpeg/Node dependencies to install yourself; use a portrait you are authorized to use — the tool does not manage rights, you do; the voice and lip-sync stages bill through whichever provider the run selects, so price per video before batch production. Against story-to-handdrawn-video on this site, the two cover opposite aesthetics: a lifelike presenter here, an illustrated flipbook there — choose by content tone.