anymouschina/TapCanvas

TapCanvas – Community Edition

What it is

• TapCanvas is an open‑source, web‑based “infinite canvas” for creating multimodal storytelling projects (film, animation, comics, etc.).
• The canvas stores scripts, characters, scene assets, storyboards, audio, subtitles and the full production workflow as typed nodes and connections.
• AI Agents read the current canvas state, plan what to do, call specialised Skills (text generation, image/video synthesis, editing, etc.) and record every step, producing verifiable deliverables.

Core capabilities

Feature	What it does
Infinite typed canvas	Nodes, ports and edges let you organise narrative elements, media assets and workflow steps in a single, version‑controlled graph.
Agent‑native execution	A unified POST /public/agents/chat endpoint routes requests through a DeepSeek Harness bridge; agents perform semantic understanding, planning, invoke skills, spawn sub‑agents and self‑validate results.
Production backbone	Supports chapter‑level context, structured storyboards, continuous shots, frame‑to‑frame hand‑off and video generation/composition (one‑click “make‑movie” workflow is in progress).
Multi‑model gateway	The Lu‑Ban API (apps/new-api) abstracts text, image, video and audio models from many providers (HeyRoute, Factory/Droid, etc.) and handles metering, pricing and dynamic model selection.
Project‑level asset management	Asset library, versioning, generation history, task status and delivery evidence are persisted; successful assets are never discarded.
Workflow & collaboration	Re‑usable DAG definitions, async workers, failure recovery, sharing/publishing, team scopes and real‑time canvas sync.

Typical use‑case flow

1. Create a project – add script, characters, scene assets and a storyboard on the canvas.
2. Define a production DAG – connect nodes that represent text generation, image creation, video stitching, subtitle generation, etc.
3. Run the Agent – the Agent reads the canvas, decides which skills to invoke, calls the appropriate model via Lu‑Ban, and stores the generated media back on the canvas.
4. Iterate & collaborate – teammates can edit the graph, view execution logs, and re‑run failed steps without losing prior work.
5. Export – final video (or other media) can be downloaded; the whole provenance graph serves as a reproducible record.

Architecture at a glance

• apps/web – React UI built with Mantine and React‑Flow; renders the infinite canvas and editor.
• apps/hono-api – Hono server exposing the REST/GraphQL API, authentication, task ledger and delivery evidence.
• apps/agents‑cli – DeepSeek Harness bridge that turns HTTP chat requests into Agent loops.
• apps/new-api – Lu‑Ban model gateway (channels for HeyRoute, Factory/Droid, etc.) handling metering and model selection.
• packages/schemas – Shared TypeScript contracts between front‑ and back‑end.
• Docker‑compose – All services (web, API, gateway, workers, PostgreSQL) run together; a single ./scripts/dev.sh docker command boots the whole stack.

Getting started (Docker)

git clone https://github.com/anymouschina/TapCanvas.git
cd TapCanvas
./scripts/dev.sh docker   # builds and starts all containers

• The script generates a one‑time OpenSSL secret for JWT/worker tokens and writes it to apps/hono-api/.env (git‑ignored).
• After it finishes, open:
  • UI: http://localhost:5175
  • API: http://localhost:8788
  • Model‑gateway admin: http://localhost:4455
• Configure your model provider keys in the Lu‑Ban admin UI (HeyRoute, Factory/Droid, etc.).
• For local development without Docker, install Node 22+ and pnpm 10.8.1, then run pnpm -w install && pnpm dev:web.

Licensing

• The core TapCanvas code is released under the MIT License.
• apps/new-api (the model gateway) retains its upstream AGPL‑3.0 license.
• All other third‑party dependencies keep their original licenses.

Who might find this useful?

• Indie filmmakers, comic creators, game writers, or any team that wants a reproducible, AI‑augmented pipeline for multimodal storytelling.
• Researchers exploring long‑chain AI workflows, provenance tracking, or agent‑driven content generation.
• Developers looking for a reference implementation of an AI‑agent‑centric production stack.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.