The creative-confidence video skill
v0.28.0 Creative briefs · beat-level QA · fail-before-render release gatesA skill your AI agent reads — Claude Code, Codex, Cursor, Kimi Code — that turns prompts, scripts, agent-readable sources, and real product flows into narrated, captioned video. 2D HTML/CSS, 3D Three.js/WebGL, and SVG motion all driven by the voice track. Local-first speech and rendering; optional tools only when the project needs them.
MIT licensed · local-first by default · cloud only when you register it
This 30-second reel was made by narova, about narova.
Speech drives everything
Start with local Piper, XTTS, Qwen, or cloned Chatterbox voices. Add an explicitly registered provider only when a project calls for it. One narrator, a full panel, or no voice at all. Give each speaker a color; the captions follow.
Every word lights up exactly as it's spoken, in the speaker's color. No manual timing, ever.
data-cue="2" keeps an element hidden until turn 2 is spoken. Visuals land on the beat.
Keep the full browser canvas and Studio in HyperFrames. Switch to Narova No-Browser when a machine cannot launch a browser: Skia draws the portable scene tree, FFmpeg plays local media and encodes the result, and unsupported HTML fails clearly instead of degrading silently.
Every scene carries a content hash. Both renderers now reuse untouched scenes from cache — changed scenes re-render, the rest are spliced via ffmpeg. "Try five versions of scene 4" costs five scene renders, not five full videos.
For ambitious work, creative-brief.md turns medium-neutral intent into 2–3 small proof branches with rationale. The agent renders decisive, project-bound evidence, rejects weak directions, restores one winner with its proof-time overrides, records exact expansion lineage, and expands only that branch. Camera, depth, and lighting stay conditional on the chosen medium.
Zero-style scenes have no implicit centering, max-width, gutter, or caption reserve. Patterns, chrome, and safeLayout are independent opt-ins. Captions remain a default-on overlay and can be disabled with captions: false.
Your agent reads product sites, articles, papers, docs, and repos, then selects the story and evidence. Narova handles the mechanical ingest and video workflow.
Explore an app in an agent-readable browser, then record highlighted semantic clicks, typing, waits, and scrolling on the narration clock. Narova frames the real UI, adds voice and word-synced captions, and refuses stale takes.
83 seconds · one continuous browser take · 24 narration-timed operations · voiced click-ripple proof
narova plan reads the canonical manifest and tells you whether a change needs TTS, alignment, mixing, composition, or rendering.
No-browser ۔ and ؟ punctuation now drive sentence timing correctly. For meaningful Urdu scripts, Narova can hand dialogue polishing to the optional urdu-voice-director skill.
Already have a recording? Set narration.file and drop in your own voice. TTS is skipped. Add wordTimings for automatic karaoke overlays — every word lights up without a single caption line in your HTML.
Use Studio or a no-browser draft plus shots --beats to inspect every promised visual beat, --reuse for revisions, and build --release for a preflight before synthesis plus a measured-timing recheck before rendering,
then export presets for TikTok, Reels, Shorts, LinkedIn, X, YouTube, and WhatsApp.
Optional by architecture
Narova ships with four local backends and keeps them as the default. Premium cloud TTS lives in separate companion skills, with no provider SDK, endpoint, model, authentication rule, or dependency inside Narova core.
API keys stay in environment variables. Provider options remain opaque to Narova and still participate in deterministic cache invalidation.
narova providers add …
No directory scanning or automatic execution.
narova-tts-provider/v1
Versioned JSON Lines over a persistent worker.
synthesize(voice, text)
Narova keeps caching, timing, captions, and rendering.
From idea to mp4
Tell your agent what you want to make. Add a URL, repo, document, script, or just an idea.
For ambitious work, your agent saves 2–3 small, orthogonal proofs with rationale, compares the rendered evidence, and expands only one selected branch. For demos, it also explores the product and records declared actions after narration timing is known.
Inspect narration and marker beats, watch Studio, ask for changes in your own words, then run the fail-before-render release gate and export final deliverables.
Install the skill once. After that, use Narova the way it was designed: talk to your agent. Pick a starting point or say it your way.
# install once $ npx skills add ammar-hasan/narova --skill narova -g # idea → video you → Make a 45-second vertical explainer about why starting small makes habits easier to keep. Make it warm and practical, and show me a preview before rendering. agent → I’ll use one warm narrator, a calm visual rhythm, and simple step-by-step scenes. I’ll keep the language practical and show you the vertical preview first.
Want direct control? The same open project works from the CLI. See the manual workflow →