VideoFree
INFINITE CANVAS · IN DEVELOPMENT

Infinite Canvas In development · ships with the site launch

On an infinite canvas, connect image, video, speech, agent, and resource processing into one auto-executing production pipeline: drop nodes, wire ports, set parameters, run. The canvas itself is a plugin app, nodes are plugins too, and existing generation capabilities plug in directly — the plugin platform grows one layer further while the framework stays thin.

THREE-LAYER DESIGN

The canvas is an app, nodes are plugins, capabilities plug in

Neither canvas editing nor node business logic enters the framework. The framework only adds generic node discovery, node invocation, and safe asset import protocols.

The canvas is a plugin app

Infinite Canvas registers as an application plugin, owning canvas editing, workflow storage, import/export, and execution scheduling. It is not built into the framework — removing it never affects generation.

Nodes are plugins too

A new canvas_node plugin type ships agent, resource, and control-logic nodes. Canvas nodes appear only in the canvas node panel — never in capability pickers, job toolbars, or the MCP tool list.

Existing capabilities plug in

Registered image, video, and audio capabilities are wrapped automatically as generic generation nodes, running through the framework's schema validation, queues, and project archiving — no per-model node code.

NODE ECOSYSTEM

Four node families cover a full pipeline

Port types are stable and versioned (text, JSON, image, video, audio, job, and more); connections are type-checked, never silently converted.

Core control

Text, number, and boolean inputs; text templates, joins, and splits; JSON parsing and structure validation; conditional branches, merges, and parallel branches.

Resource import

Browse and pick images, video, and audio from the project library; upload from your computer; import public external resources; write results back into the project library.

Agent nodes

Structured script and copy writing, storyboard planning, prompt review, and constrained AI decisions. Outputs must pass JSON Schema validation, and callable capabilities are declared by allowlist.

Generation nodes

Pick a capability_id, wire the prompt port and declared media inputs, set parameters and defaults — reusing existing plugins without copying any generation code.

WORKFLOW SHARING

Workflows are JSON — take them anywhere

The canonical carrier of a workflow is a JSON document: the canvas saves it as a project document and supports standalone import and export.

  • Share graphs alone as *.vfworkflow.json — graph structure and asset references, no actual media; optionally bundle small fixed assets in a .vfworkflow package.
  • Shared files never contain API keys, system tokens, absolute paths, or temporary upload credentials; assets keep only logical references, type, name, and SHA256.
  • Import validates format version, node and port existence, required plugins and version compatibility, connection types, required inputs, and cycles; missing assets are rebound by the user.
  • A missing plugin triggers an install hint — never an automatic download or code execution.
EXECUTION & RECOVERY

Failures never start you over

Workflow definitions and individual runs are stored separately — a run is a first-class, recoverable object.

  • The scheduler runs in topological order; independent branches execute in parallel under a concurrency cap.
  • A failed node stops its downstream by default while successful outputs are kept; retry reruns only the failed node and its downstream.
  • Only nodes declared idempotent may auto-retry; generation jobs can be genuinely cancelled.
  • Run records snapshot plugin versions, node versions, and the workflow itself, so plugin updates never silently change old pipelines.
Safety boundaries: no arbitrary-code nodes and no unbounded loops in the first version; external URL import is strictly SSRF-guarded (internal and cloud-metadata addresses rejected, DNS and redirects validated, protocol, size, and time limited); AI never batch-launches expensive jobs without confirmation.

Explore the plugin platform today

The canvas ships with the next release. Build your capabilities as plugins now, and they become nodes the day it lands.

Download VideoFree → Plugin guide