Narova — open video production system for humans and agents

Narova is an open, local-first video production system for humans and agents. It turns creative intent — a prompt, script, scene file, source, repository, or real product flow — into directable, reproducible video: structured scenes, deterministic timelines, local speech, word-synced captions, multiple local renderers, AI clip generation, real product walkthroughs, revisions, provenance, and release.

The CLI and the Narova agent skill are separate artifacts. Installing this npm package adds the narova, narova-setup, and narova-uninstall commands; it does not install or modify agent instructions.

What you build and publish with Narova is your choice, and you are fully answerable for it — under law, by your own ethics, values, and conscience, and by the religious or cultural commitments you hold. Narova gives you inspectable sources, reviewable creative proofs, and explicit provider choice to support that judgment; it does not verify legality, rights, or acceptability for you.

The skill pins one compatible CLI release. After the skill is updated, its next session checks narova --version and reconciles this global package to that exact version before use.

Website · npm package · GitHub · Machine protocol · Agent skill · Issues

Install

Narova supports macOS and Linux. Windows users should run it through WSL. Node.js 18+, Python 3.10+, FFmpeg, and FFprobe are required.

npm install --global @narova/narova
narova doctor

The default local Piper voice environment is created on the first synthesis, or explicitly with:

narova-setup

Optional --xtts, --qwen, and --chatterbox flags install larger local voice backends into Narova-owned virtual environments. Optional hosted speech (ElevenLabs, OpenAI) and video generation (Sora through OpenAI, Runway) are separate companion skills registered explicitly; the core package stays local-first and contains no vendor API adapter.

Quick start

npx @narova/narova demo    # one command to a finished MP4
narova init my-video
cd my-video
narova check
narova build --release
narova provenance

Projects can be exchanged as deterministic, digest-verified .narova files:

narova pack --project my-video --output my-video.narova
narova open my-video.narova --inspect
narova remix github:owner/repository#main --dir my-remix

See the archive compatibility profile for the format, bounds, trust notice, and extraction rules.

What it makes

• Prompt-to-video, script-to-video, and source-grounded explainers
• Narrated dialogue with local TTS and word-synced captions
• Silent and marker-driven motion pieces (narration is optional)
• Deterministic 2D HTML/CSS/SVG and Three.js/WebGL scenes
• AI clip generation through explicitly selected providers (Sora, Runway), with optional creator-owned character/object/place continuity
• Real product walkthrough videos with timed browser actions
• Local MP4, SRT, and VTT deliverables without a render service
• Deterministic shareable project archives with safe inspection and remix lineage
• Read-only evidence-graded provenance reports and text, YouTube, web, or JSON credit output

Two local renderers ship with the package. HyperFrames is the full browser canvas (HTML/CSS, WebGL, Studio); No-Browser draws a portable scene tree with Skia when a machine cannot launch a browser. Both run locally with no render service or fee.

Agents can consume the versioned narova.result/1 machine protocol (--json on every operation) with stable exit classes, diagnostics, and artifact records — no parsing of terminal prose.

For generated shots, an optional continuity block in reel.config.* names entities and per-shot keep/change intent. Select it with narova generate "<prompt>" --continuity <shot-id>. A shot may include one local image anchor only when the selected provider declares reference-image support; otherwise the command fails before provider work. Recipe version 3 records the exact selected context and anchor identity for inspection and regeneration. Narova does not infer entities, choose style/camera, rank results, or certify provider pixel adherence.

See the project README for the scene-script format, renderer choices, product walkthroughs, source grounding, proof branches, judge/assertions, and full workflow.

Agent skill

Install the separate agent skill with:

npx skills add ammar-hasan/narova --skill narova -g

Network and local data

Narova renders locally. First use can download the pinned HyperFrames CLI, Python packages, and selected speech models. Optional stock providers, AI clip generation, browser capture, and external voice providers use the network only when explicitly selected. Models, caches, saved voices, and provider manifests live under ~/.narova by default and are retained across CLI upgrades.

Update or remove

npm install --global @narova/narova@latest
npm uninstall --global @narova/narova

Updating the npm package alone does not update agent instructions. Update the skill with npx skills update narova -g; its next session enforces the matching CLI version.

Releases from 0.31.1 onward use npm Trusted Publishing and include provenance linking the package to its public GitHub source and publishing workflow. The manually bootstrapped 0.31.0 release has no provenance attestation. Narova is available under the Apache-2.0 license.