anymouschina/TapCanvas

What it solves

TapCanvas provides a unified, traceable workspace for AI-driven multimedia production. It solves the fragmentation of the creative process by bringing scripts, character and scene assets, storyboards, images, video, and audio into a single infinite canvas workflow, moving beyond simple one-off model calls to a structured production pipeline.

How it works

The system operates as an "Agent-native" canvas where AI agents read the project state and autonomously select skills and tools to execute multi-step production tasks. The architecture consists of several key components:

• Infinite Canvas: Uses strongly typed nodes and connections to organize assets and workflows.
• Agents Bridge: A gateway (based on DeepSeek Harness) that handles semantic understanding, planning, and delivery self-inspection.
• Production Workflow Kernel: Manages the execution of tasks via asynchronous workers, ensuring that long-chain production is recoverable and traceable through task ledgers and delivery evidence.
• Multi-Model Gateway: A unified API (based on new-api) that manages access, metering, and dynamic dispatch for text, image, video, and audio models.
• Director's Desk: An integrated 3D scene builder for character positioning and camera pre-visualization.

Who it’s for

It is designed for creators producing AI-driven films, comic-style dramas, and multimodal content who need a structured way to manage complex assets and multi-step AI generation workflows.

Highlights

• Agent-Native Execution: Agents autonomously plan and execute production steps based on the canvas state.
• Traceable Workflow: Every asset is linked to its generation history and delivery evidence, preventing loss of successful outputs.
• Infinite Canvas Interface: A visual DAG (Directed Acyclic Graph) approach to organizing multimodal assets.
• 3D Pre-visualization: Integrated tools for 3D scene setup and camera movement planning.
• Unified Model Access: A single gateway to manage dozens of different LLMs and generative media models.