Skip to content
← Back to Demos
Content & Design

AI Story Studio

Turn a dataset, article URL, or research topic into an animated multi-scene data story with 18 scene templates (charts + typography), narration, cross-model verification, exportable as ZIP / PPTX / browser-side MP4. Long-form mode runs an agentic outline → viz-critic audit → parallel chapter-expand pipeline for explainer-length pieces. PREMIUM tier (v2.0) adds server-side 1080p60 Remotion rendering for buttery, broadcast-grade output. The 3Blue1Brown / Vox / Bloomberg treatment for any numbers you throw at it.

AI ContentData VizStorytellingPresentation
Share

Demo preview — limited to 3 uses per day. Sign up free for 1 use per tool per day + 15 bonus tokens, or try the full tool.

Like this demo? Get the full tool.

1 free use per day. No credit card required.

Use the Tool→

Overview

AI Story Studio converts raw data into an animated, narrated, shareable story — the kind content creators use to make a trend land with a wide audience. Three input modes cover the common use cases: paste a dataset (CSV / TSV / JSON), paste an article URL (we extract the quantitative content), or type a topic for AI-powered research. The output is a multi-scene Vega-Lite story with narration beats, playable in-browser, screen-recordable for video work, and shareable as a public link.

How It Works

  1. Choose an input mode. Paste Data is available in every tier; URL and AI Research unlock with Advanced.
  2. Provide the source. CSV/TSV/JSON paste, article URL, or a topic + provider pick (Perplexity for search+synthesis in one call, Tavily for raw search with our composer synthesizing). Optional hint lets you steer the angle.
  3. We compose. A primary model (gpt-4o-mini in Basic, gpt-4.1 or claude-sonnet-4 in Advanced) identifies the most compelling insight, builds a narrative arc of 3–5 scenes (Basic) or 3–15 scenes (Advanced), and emits a Vega-Lite chart spec + narration for each.
  4. Advanced: cross-model verification. A second model (opposite vendor from the primary — Claude audits GPT and vice versa) fact-checks every scene against the source. You see a transparent confidence score and a flag panel with type (unsupported claim / number mismatch / contradiction / missing source), the exact excerpt, the verifier's reasoning, and an optional suggested fix. Accept fixes to re-render, or ignore them — you stay in control.
  5. Play or share. Auto-play runs the full story with Framer Motion transitions between scenes (fade / slide / morph). Space pauses, arrows scrub. Share Link saves to Supabase and returns a public URL with a 12-char token — opens as an immersive viewer with source attribution.

What You Can Do With It

  • Turn your quarterly sales CSV into a 30-second narrated animation for the board deck.
  • Paste a news article about EV adoption and get a properly-sourced visual breakdown, not a paraphrase.
  • Type "the shift to remote work, 2020–2025" and get a Perplexity-researched, citation-linked, cross-verified mini-doc.
  • Screen-record any story with QuickTime / OBS / Premiere to drop into YouTube or a pitch video.
  • Share the link in a Slack thread — viewers see the same interactive player, not a screenshot.

Pricing

  • Basic — 1 token. Paste data mode. 5 scenes max. gpt-4o-mini. All 18 scene templates (12 chart + 6 display, including PIE_CHART, FACETED_PIE, CHOROPLETH_MAP, SCATTER, DUMBBELL, HEATMAP added in v1.4). Outline-stage template-selection rubric + variety rule (no template > 2x per deck) baked into the prompt. Single-pass composition.
  • Advanced — 4 tokens. All three input modes (paste, URL, AI research). 15 scenes max. Premium model picker (GPT-4.1 / Claude Sonnet 4). All 18 scene templates. Cross-model verification with multi-option fix panel. Per-scene regeneration. ZIP bundle with animated WebM clips per display scene + static PNGs (v1.2.4). Full-story MP4 export via in-browser ffmpeg-wasm — single playable H.264 file ready for Premiere / PowerPoint embed / YouTube (v1.3). PowerPoint deck with real OOXML slide transitions + per-shape fadeIn entrance animations (v1.2.3). Custom branding on shared pages. Single-pass composition (one LLM call).
  • Long-form (Advanced + sub-toggle) — 7 tokens. Same as Advanced, plus flips composition to a multi-step agentic pipeline: an outline call commits 3-6 chapters with strict per-chapter scene counts, a per-scene viz-critic LLM call then audits the outline's template choices vs your dataPoints (proposing targeted swaps when a chart doesn't fit the data shape — e.g., HIGHLIGHTED_TREND on ranking data without a time axis gets swapped to HORIZONTAL_BAR_RANKED), then chapter-expand calls run in parallel (each with full context budget) to deliver up to 20 scenes worth of structurally coherent long-form content. Best for "explainer-length" pieces where single-pass loses the thread. ~30-60s wall time per story. Single-pass fallback per chapter on timeout, so the story always ships.
  • Premium (Advanced + sub-toggle) — 14 tokens. Everything in Long-form (agentic pipeline + viz critic + verification), plus server-side Remotion rendering that produces a true broadcast-grade 1080p60 MP4 — frame-driven motion timing instead of html-to-image's ~12 fps ceiling. The render runs on our infrastructure (no client-side ffmpeg-wasm), streams progress to the widget over a single fetch (no polling), and lands in your account permanently as a public share URL. Foundation for v2.0.1+ TTS narration, brand kit, per-scene editing, and the designer typography pass. ElevenLabs voice picker (when v2.0.1 ships) adds +4 tokens (18 total).

Audiences

  • Business — executives, analysts, consultants who need to explain numbers to non-technical audiences.
  • Education — teachers and course builders who want data-driven explainers without custom D3 work.
  • Entertainment — content creators and journalists producing YouTube / TikTok / newsletter segments.

Roadmap

  • v1.1 — Creative output overhaul (shipped 2026-04-25): 7 scene templates, 8 topic-aware palettes, ~2000 Lucide icons, Framer Motion intra-scene animations, ZIP + PPTX export.
  • v1.1.1 — Hotfix (shipped 2026-04-25): count-up animation, Vega-Lite v5 → v6 schema URLs, icon 404 onError fallback, chapterLabel discipline for the deck-cutoff bug.
  • v1.2 — CREATIVE tier with agentic generation pipeline (shipped 2026-04-25): outline → parallel chapter-expand → stitch → verify; new creative_cost column on the tools table; 6-token Long-form sub-toggle inside ADVANCED.
  • v1.2.1–v1.2.4 — Hotfix run (shipped 2026-04-25): chart axis coercion (year-temporal → ordinal, percent format swap), highlight-annotation halo, ZIP CSP fix, PPTX dark-theme legibility, OOXML-real PPTX slide transitions + per-shape fadeIn entrance, PICTOGRAPH layout rebuild (quantitative scale, dynamic icon sizing, count-text halo), [object Object] title fix, animated WebM clip per display scene in ZIP exports.
  • v1.3 — Visualization quality + animated outputs (shipped 2026-04-25): 5 new scene templates (HORIZONTAL_BAR_RANKED, STACKED_BAR, GROUPED_BAR, STAT_GRID, QUOTE_CARD); outline-stage template-selection rubric across all tiers; per-scene viz-critic LLM call (Long-form tier — single-pass audit of chart choices vs your data points, applies targeted swaps); full-story MP4 export via in-browser ffmpeg-wasm; Twemoji colored-emoji support via tw: prefix routing through jsdelivr CDN. Long-form cost adjusted 6 → 7 tokens to cover the viz-critic call.
  • v1.3.1 — Hotfix (shipped 2026-04-25): blob: added to CSP connect-src so MP4 export via ffmpeg-wasm works; HORIZONTAL_BAR_RANKED sort warnings (sort:"-x" → sort:null on pre-sorted data); PPTX per-shape fadeIn animations now actually emit (extractShapeIds regex was attribute-order dependent, missed half the shapes).
  • v1.4 — More viz variety + smarter critic (shipped 2026-04-25): 6 new chart templates (PIE_CHART donut, FACETED_PIE for percentages-across-segments, CHOROPLETH_MAP world + US states, SCATTER with optional regression line, DUMBBELL for paired before/after, HEATMAP for 2D matrices). Outline + viz-critic prompts now include an explicit "no template > 2x per deck" variety rule, 11 anti-pattern checks, and 4 worked-example swaps the critic learns from. Total templates: 18 (12 chart + 6 display).
  • v2.0.0 — PREMIUM tier with server-side Remotion rendering (shipped 2026-04-25): new 4th tier at 14 tokens. Frame-driven 1080p60 MP4 rendered on our infrastructure (not your browser). All 18 scene templates ported to Remotion compositions. Pre-rendered Vega specs via vega.View.toSVG() (no node-canvas). Streamed NDJSON progress over a single fetch — no client polling. New render_jobs Supabase table + story-renders Storage bucket for permanent share URLs. RenderProvider interface designed for a future Remotion Lambda swap.
  • v2.0.1 — Coming next — TTS audio narration baked into the MP4. OpenAI default; ElevenLabs voices as a +4-token surcharge (18 total). Voice picker UI inside the widget.
  • v2.0.2 — Brand kit: logo, accent color, custom font upload (woff2), intro/outro stings. Saved per user/org.
  • v2.0.3 — Per-scene editing UI: drag-handle reorder, narration editor, alternate-chart picker, trim duration, partial re-render.
  • v2.0.4 — Designer typography pass: font hierarchy, motion curves, letter-spacing/leading per template (3Blue1Brown / Vox / Bloomberg reference targets).