anymouschina/TapCanvas

TapCanvas Pro – AI‑powered visual content‑creation canvas

What it is – TapCanvas Pro is an open‑source, monorepo‑structured web application that lets you build long‑form AI‑generated media (text, images, video, story‑boards) on a single, infinite canvas. It ties together a front‑end canvas, a back‑end API, and an Agents execution layer so that assets, prompts, and model calls are persisted as part of a project and can be reused across steps.

Core components

Component	Tech stack	Role
apps/web	Vite + React 18 + Mantine + React‑Flow	Interactive canvas UI where nodes represent text, images, video, scripts, etc. Drag‑and‑drop assets and connect them to define a workflow.
apps/hono-api	NestJS + Hono + Node.js + Prisma + PostgreSQL + Redis	HTTP API that stores projects, assets, authentication, and exposes AI‑tool schemas.
apps/agents‑cli	TypeScript	“Agents Bridge” – a minimal autonomous‑agent loop (plan → tool → report). Handles semantic understanding, skill selection, memory, sub‑agents and forwards calls to the configured LLM endpoint.
apps/new‑api	Modified new‑api gateway	Unified gateway for multiple model providers (text, image, video). Handles routing, quota, and billing.
packages/*	Shared TypeScript packages	Schemas, SDK, reusable utilities used by both front‑end and back‑end.

Key features

• Infinite canvas workflow – nodes for documents, characters, story‑boards, generated images/video, and execution links can be iterated indefinitely.
• Agents‑driven execution – a POST /public/agents/chat endpoint routes user prompts to the Agents Bridge, which decides which tool (e.g., text generation, image synthesis) to call and writes the result back to the canvas.
• Multi‑model integration – the new‑api gateway lets you plug in any supported LLM, diffusion model, or video model behind a common schema.
• Project‑level asset persistence – role cards, reference images, script fragments, and execution logs are stored per project, enabling continuity across sessions.
• Chapter‑wise creation – designed for long‑chain scenarios such as novel‑to‑storyboard or episodic video production, with context injection scoped to a chapter.
• Docker‑first deployment – a ready‑made docker‑compose.yml spins up PostgreSQL, Redis, the agents bridge, the API, and the model gateway with a single command.

Typical use cases

• Turning a novel or script into a storyboard and then into a short video.
• Collaborative teams that need to manage prompts, generated assets, and execution history in one place.
• Any workflow that requires several AI steps (e.g., generate a scene description → create concept art → synthesize a video clip) with traceable results.

Getting started

1. Choose a launch method – either the one‑click Docker Compose flow (recommended for quick try‑outs) or a local source‑code setup for development.
2. Install dependencies – the repo uses a pnpm workspace (pnpm -w install).
3. Configure – copy the .env.example files for the API, web front‑end, and agents‑cli, then set at least DATABASE_URL, JWT_SECRET, and AGENTS_BRIDGE_BASE_URL.
4. Run –
   • Docker: cd apps/hono-api && docker compose up -d then pnpm dev:web.
   • Source: start three terminals: pnpm --filter agents dev -- serve --port 8799, pnpm dev:api, and pnpm dev:web.
5. Open http://localhost:5173 in a browser; the canvas UI should load and be able to talk to the API and agents bridge.

License – The community edition of TapCanvas is released under the MIT License. It is explicitly separated from the commercial version; the MIT terms apply only to the code in this repository.

---

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