Flovart

The next direction is native Flovart effects inside creative software: generate a version, refine it in the host, and open Workflow for complex work. Windows AE/PR effects come first; macOS will be evaluated separately. Native effects remain planned, distinct from today's experimental panels. See the product and implementation design.

See the workspace

The current screenshots are real Workflow and Skill surfaces. A future recording that shows an agent changing that same Workflow is specified in README_VISUAL_TODO.md; this README does not fabricate that evidence.

Why Flovart?

Most AI creative tools make you choose between a visual editor and an autonomous agent. Flovart is built around one production state that both can use:

• Agent-native: a coding agent reaches the visible Workflow through typed Flovart Link and CLI operations, not mouse automation or a hidden project copy.
• Human-editable: nodes, connections, assets and results stay in the Workflow you can inspect and continue editing.
• BYOK and multi-provider: bring your own model services and API keys; real provider and billing behavior remains explicitly tracked as Experimental where it is not certified.
• Production Skills: package visual language, shot rules, workflow recipes, checkpoints and acceptance criteria into reusable production methods.

One workspace, two ways to create

With your agent

Creative brief
      ↓
Codex / Claude Code / OpenCode
      ↓
Flovart Link + CLI
      ↓
Inspect · Apply · Run
      ↓
The visible Workflow

For example:

“Open Flovart and build a three-shot product video Workflow from these references.”

The agent reads the current project and revision, applies explicit operations, and can run a confirmed node. The exact Host status is shown in the Support Matrix.

By hand

You can still add, move, resize and connect nodes, drop local image or video files, configure a model, run generation, inspect artifacts and iterate visually. There is no separate “agent version” of your project: the human and agent paths converge on the same Workflow authority.

What makes Flovart different?

What you can make

• Agent-native Workflow: compose image, text, video, audio and configuration nodes into visual generation flows.
• Image and video generation: choose modes, references and parameters, then monitor results and recovery through the Workflow.
• Reference-aware production: combine graph references, mentions, assets and generated artifacts as typed inputs instead of pasting everything into one prompt.
• Production Skills: reuse a method for a visual language, shot structure or production checklist across projects.
• Local projects and assets: keep projects, references and generation history close to the workspace, with no cloud-sync promise.
• Extensible contracts: providers, hosts, node operations and Skills meet explicit contracts so integrations can be tested without pretending a mock is a certified host.

The product surfaces behind this story are Workflow for generation orchestration, Table for the still-evolving media-preprocessing workbench, and Agent for the spatial production-control surface. Table and Agent are real application entries, but their remaining implementation and migration work is not presented as complete; see Features.

Production Skills

Prompts are reusable text.

Production Skills are reusable production methods. A Skill can capture:

• visual language and style rules;
• shot structure and workflow recipes;
• checkpoints and human approvals;
• model policy, cost boundaries and safety rules;
• acceptance criteria for the final artifact.

The repository contains a local Flovart Skill surface and the VOX Skill reference. The broader community contract and ecosystem are still in design and implementation, so the Skill section is an active capability rather than a promise of a mature marketplace. Start with the Skill guide.

Bring your own models

Your provider
      ↓
Your API key
      ↓
Your assets + Workflow
      ↓
Your generated result

Flovart does not bundle model services. Configure a provider in the app, choose the capabilities and model you need, and keep the provider terms, cost and output rights in your own hands. OpenAI-compatible BYOK and remote-provider paths are currently Experimental; adapter presence is not paid-provider certification. Use the Support Matrix as the status source of truth.

Compatibility

Experimental, Planned and External Gate items are not Stable claims. Evidence, boundaries and release gates live in the Support Matrix; this table is intentionally not a second compatibility policy.

Quick start

For creators

1. Download a preview build from GitHub Releases.
2. Open Flovart and add an AI service in Settings.
3. Create or open a Workflow, add references, and start creating.

The public Releases page may contain test or preview artifacts; it is not a claim that every Host or Provider is Stable.

For coding-agent users

Use the repository-verified source path while the versioned CLI package publication remains a release gate:

git clone https://github.com/avabbbb/Flovart.git
cd Flovart
npm install
npm run flovart:cli -- start --source --web --open
npm run flovart:cli -- workflow.inspect --json

Then ask your local agent: “Open Flovart.” The normal agent loop is status, workflow.inspect, workflow.selection.get, workflow.apply and workflow.node.run; the Getting Started guide covers development-only diagnostics and isolated browser checks.

Architecture

flowchart LR
  H["Human creator"] <--> W["Live Workflow"]
  A["Coding-agent harness"] --> L["Flovart Link + CLI"]
  L --> W
  W --> P["Provider adapters"]
  W --> R["Local assets + a