Beatboard

Beatboard turns a comedy idea into a shoot-ready storyboard with timed beats, generated art, voiced dialogue, direction, and dependency-aware revisions in minutes.

Comment

Inspiration

The hardest part of making a short comedy often happens before anyone presses record. You need a clean setup, a turn that feels earned, a punchline with the right timing, and enough direction for performers to know how to play it.

Most generative video tools jump straight to the finished output. That can hide the creative decisions underneath and make small changes expensive.

We built Beatboard as an agentic pre-production studio that turns a simple comedy idea into something creators can understand, rehearse, revise, and take into a real shoot.

Beatboard is submitted to The Taskmaster track of All Things Agentic and the ClickHouse track of Agentic Cinema.

What it does

Beatboard turns three inputs, a premise, target runtime, and comedy style, into a timed interactive storyboard.

The agent plans the cast, setting, dialogue, actions, camera framing, performance direction, timing, and relationships between beats. It then generates storyboard frames, gives the characters distinct voices, and assembles everything into a playable animatic.

Creators can view the same scene through four directorial lenses:

  • Storyboard shows what happens.
  • Context shows what each character knows.
  • Subtext shows what each character really means or wants.
  • Supertext shows what the audience is meant to understand and why the comedy works.

Beatboard also supports dependency-aware revisions. If one beat changes, the agent determines the smallest affected set instead of blindly rebuilding the whole storyboard.

When the board is ready, Beatboard exports a Shoot Pack PDF containing storyboard frames, dialogue, timing, camera cues, performance notes, and cast information for use during production.

How we built it

Beatboard uses a React + Vite frontend and a Python FastAPI backend deployed on Google Cloud Run.

The core planning workflow uses Gemini 3.6 Flash with the Google Gen AI SDK to turn a brief into a structured beat graph containing characters, dialogue, timing, production cues, and dependencies. This satisfies the All Things Agentic requirement for Gemini 3.5+ and a Google agent framework, while Cloud Run provides the required Google Cloud infrastructure.

For Agentic Cinema, the agentic workflow is also integrated with Google Cloud Agent Builder / Gemini Enterprise Agent Platform and ClickHouse as the partner runtime service. ClickHouse is not passive logging. Beatboard reads and writes production state at runtime, including comedy style grammars, active beats, dependencies, and revision history, and that stored state affects subsequent planning and revision decisions.

Google image generation creates the storyboard panels, Google Cloud Text-to-Speech gives each character a distinct voice, and Google Cloud Storage holds generated media assets.

Together, these services form an end-to-end agentic workflow:

brief → plan → production memory → storyboard → voiced animatic → dependency-aware revision → shoot-ready handoff

Challenges we ran into

The biggest challenge was keeping one coherent production across several nondeterministic AI services.

Characters and locations had to remain recognizable across separately generated storyboard frames. Dialogue also had to fit a real 15- or 30-second runtime without unnaturally speeding up the voices. We solved that by measuring generated TTS duration and shortening the script when necessary.

We also needed Gemini output to be reliable enough to drive an application, so we combined structured schemas and probabilistic creative reasoning with deterministic timing, caching, assembly, and graceful fallbacks.

Dependency-aware revision created another challenge. Changing one beat should not automatically regenerate everything. We modeled setups, punchlines, callbacks, and other relationships explicitly so Beatboard can query production memory and determine the smallest affected set.

Accomplishments that we're proud of

We're proud that Beatboard became a complete workflow rather than a chatbot or collection of disconnected AI features.

A short brief becomes a structured, illustrated, voiced, timed, and playable production plan. Two performers can hear the scene, understand how to play it, inspect its hidden layers, and leave with a practical Shoot Pack they can use during production.

We're especially proud of the Context, Subtext, and Supertext lenses, which expose different layers of the same scene without changing the storyboard itself.

We also built natural-duration dialogue, distinct character voices, dependency-aware revisions, persistent ClickHouse production memory, resilient media generation, Cloud Run deployment, and a one-click PDF production handoff.

What we learned

We learned that agentic media production does not have to mean generating every final frame.

There is substantial creative leverage in pre-production, where an agent can take a goal and carry out a multi-step workflow that turns an idea into something humans can rehearse and execute.

We also learned that storing data is not the same as giving an agent memory. ClickHouse became meaningful when Beatboard began querying style rules, current beat state, dependencies, and revision history before deciding what to do next.

Finally, reliable agentic software needs both probabilistic and deterministic components. Gemini provides creative reasoning, while schemas, measured timing, explicit dependencies, caching, persistent state, and deterministic assembly make the workflow dependable.

What's next for Beatboard

Next, we want Beatboard to become a richer shared pre-production space without losing its simplicity.

That includes reusable character and visual bibles, real-time collaboration, alternate takes, side-by-side versions, additional comedy structures, more languages and voices, and direct handoff into common production and editing tools.

Longer term, Beatboard's production memory could learn how a particular team works and which structures, performances, and pacing choices tend to succeed for them while keeping the final creative decisions with the people making the scene.

One thing I would verify before pasting this

Your current README clearly documents Gemini, the Gen AI SDK, Cloud Run/Google Cloud services, and ClickHouse, but it does not currently make the Google Cloud Agent Builder integration explicit.

If Agent Builder is genuinely present in the runtime, update the README and architecture diagram to show it explicitly too. If it is not actually present, do not claim it in the Project Story. In that case, Beatboard still appears well aligned with All Things Agentic, but you should address the missing Agentic Cinema requirement in the implementation rather than through wording alone.

Built With

Share this project:

Updates