Open-DreamFlow Studio

Open-DreamFlow is a production-grade, MIT-licensed open-source spatial AI filmmaking workstation that combines Dreamina 2.0 Smart Canvas + Google Flow (gemini-omni-1.2-flash 15s + Scene Builder) with a Zero-Markup BYOK + 12-Tool FFmpeg 8.0 MCP Engine + Live Browser Bridge architecture.

Instead of paying 3×–8× SaaS credit markups or juggling disconnected tabs, filmmakers compose on a 95%-open #0e0e0e matte charcoal spatial canvas stocked with real 35mm camera photography presets, cyan #20D5EC Bezier wire-drop menus, True First-Frame → Last-Frame Keyframe Bridges (A → B), 64-Bar Interactive Waveform Audio & Voiceover TTS Nodes, interactive Inline Mask Inpainting / Outpainting / RGBA Cutout / 4K Lanczos Upscaling, @Elements 6-reference identity vaults, 1-click 9-Grid Contact Sheet Slicing, Smart Multiframes (2–10 keyframes), disk project persistence (.dreamflow.json bundles), and an embedded Timelines 1 Dual-Lane (V1 + A1) Sequence Node that mixes stereo audio into H.264/AAC .mp4 masters and exports Premiere Pro / Final Cut Pro .fcpxml.

System Architecture

flowchart TD
  subgraph Launcher["Unified Single-Command Launcher & Container"]
    CLI["bin/open-dreamflow.mjs (open-dreamflow start | doctor)"]
    DOC["Dockerfile + docker-compose.yml (Node 22 + FFmpeg + Chromium)"]
  end

  subgraph Studio["apps/web-studio (1:1 Dreamina Matte Charcoal Canvas · React 18 + TS + Zustand)"]
    CB["Infinite Spatial Board (#0e0e0e) + 3 Genuine 35mm Camera Presets"]
    KB["A → B Keyframe Bridge Strip + Inline Mask Inpaint / Outpaint / RGBA / 4K"]
    AU["64-Bar Scrubbable Waveform Audio Cards + Voiceover TTS / Ambient Foley"]
    TL["Timelines 1 Dual-Lane (V1 Video + A1 Audio) + Fullscreen Editor (.mp4 + .fcpxml)"]
  end

  subgraph Engine["packages/mcp-client (:3001 Single-Port SPA + Real FFmpeg 8.0 + Disk Store + WS Relay)"]
    API["REST Media Pipeline (/api/keyframe-bridge, /generate-audio, /compile-timeline, /inpaint, /outpaint, /remove-bg, /upscale)"]
    PRJ["Disk Project Store (.dreamflow-data.nosync/projects + .dreamflow.json Bundles)"]
    MCP["JSON-RPC 2.0 MCP Server (12 Callable Spatial Tools)"]
    WS["Live WebSocket Bridge Relay (ws://localhost:3001/bridge)"]
  end

  subgraph Providers["Zero-Markup Cloud BYOK & Browser Credit Routes"]
    SDK["@open-dreamflow/provider-sdk (Gemini Omni 1.2 · Seedance · Fal · Replicate · ComfyUI)"]
    EXT["extensions/chrome-bridge (Manifest V3 Popup + Content Capture for Dreamina & Flow)"]
  end

  CLI & DOC --> Engine
  Engine --> Studio
  CB & KB & AU & TL --> API & PRJ & MCP & SDK
  EXT <--> WS
  WS --> CB

Related MCP server: YoloCut

Dreamina 2.0 + Google Flow Feature Matrix

Monorepo Structure

open-dreamflow/
├── bin/
│   └── open-dreamflow.mjs      # Unified CLI (`open-dreamflow start [--port 3001]` & `open-dreamflow doctor`)
├── apps/
│   └── web-studio/             # React 18 + TypeScript + Vite + Tailwind + Zustand 1:1 Dreamina UI
│       ├── public/media/       # Genuine camera EXIF JPEGs, organic 35mm H.264 MP4s & stereo foley MP3
│       └── src/
│           ├── components/     # CanvasBoard, DreaminaFloatingChrome, ContextualNodeOverlay,
│           │                   # InlineImageEditorModal, TimelineCanvasNode, ShareableAppView, ProviderModal
│           ├── presets/        # 3 natural photography presets (Salt & Terracotta, Kyoto Joinery, Alpine Pass)
│           ├── store/          # Zustand store + Disk Auto-Save + Snapshots + Keyframe Bridge + Audio/TTS
│           ├── services/       # Zero-Markup BYOK, dynamic :3001 REST/MCP client, WS bridge client
│           └── utils/          # FCPXML v1.10 exporter, warm Kodak 250D fallbacks, URL recipe codec
├── packages/
│   ├── provider-sdk/           # @open-dreamflow/provider-sdk (Gemini Omni 1.2, Seedance, Fal, Replicate, ComfyUI)
│   └── mcp-client/             # @open-dreamflow/mcp-client (:3001 Single-Port SPA + FFmpeg 8.0 + 12 MCP Tools)
├── extensions/
│   └── chrome-bridge/          # Manifest V3 Chrome Extension + Popup syncing Dreamina & Google Flow via WS/HTTP
├── Dockerfile                  # Production single-container image (Node 22 + FFmpeg)
└── docker-compose.yml          # One-command container orchestration with persistent project/media volume

Quickstart

1. One-Command Unified Launch (Recommended)

Build once and launch the entire studio—Web Studio SPA, FFmpeg 8.0 media pipeline, 12-tool MCP server, disk project store, and WebSocket Chrome Bridge—on one single port (http://localhost:3001):

cd open-dreamflow
npm install
npm run build

# Run environment diagnostics (Node, FFmpeg, ffprobe, TTS, built assets)
npx open-dreamflow doctor

# Launch unified single-port Open-DreamFlow Studio on http://localhost:3001
npx open-dreamflow start --port 3001 --no-open

Open http://localhost:3001. Everything works out of the box with zero API keys: switch between the 3 35mm camera presets (Salt & Terracotta, Kyoto Joinery, Alpine Pass), bridge keyframes (A → B), synthesize voiceovers or stereo foley with 64-bar waveforms, paint inpaint masks, slice 9-grid contact sheets, and compile V1 + A1 .mp4 masters.

2. Docker One-Command Launch

docker compose up --build

Then open http://localhost:3001. Projects and generated media persist cleanly in the mounted .dreamflow-data.nosync volume.

(For Vite HMR frontend development, you can still run node packages/mcp-client/dist/server-bridge.js --port 3001 alongside cd apps/web-studio && npm run dev on :8789—executionRouter.ts auto-detects both single-port :3001 and split :8789 → :3001 modes.)

Run end-to-end verification tests across packages:

npm test

3. Zero-Markup BYOK Setup

Click BYOK · MCP in the top-right header (or click the model pill in the contextual prompt bar):

1. Google AI Studio (gemini-omni-1.2-flash): Paste your AIzaSy... key for direct POST /v1beta/interactions calls (4–15s video, first/last-frame keyframe conditioning, up to 6 @Element refs, and previous_interaction_id multi-turn branching) plus Gemini image editing on /api/inpaint.

2. BytePlus Ark (seedream-3-0 / seedance-1-0-pro), Fal.ai (fal_...), Replicate (r8_...), or Local ComfyUI (http://127.0.0.1:8188): Raw pass-through execution with 0% platform markup.

3. Keys persist strictly in browser localStorage (open_dreamflow_byok_v1) and can be cleared in 1 click.

4. MCP Server & Local FFmpeg 8.0 Engine (:3001)

packages/mcp-client exposes 12 real JSON-RPC 2.0 MCP tools at http://localhost:3001/mcp and ws://localhost:3001/mcp:

• generate_video, keyframe_bridge, generate_audio, flow_scene_builder

• slice_contact_sheet, extract_last_frame, inpaint_frame, outpaint_expand

• remove_background, upscale_4k, extract_style_dna, save_project

Any connected MCP tool can be spawned directly onto the spatial canvas as an executable node card from BYOK · MCP → MCP Servers (:3001).

5. Chrome Extension Subscription Bridge (ws://localhost:3001/bridge)

Capture generated frames, videos, and prompts from dreamina.capcut.com and labs.google/fx/tools/flow (flow.google.com) directly onto your Open-DreamFlow canvas:

1. Open chrome://extensions in Chrome, Arc, or Brave, enable Developer mode, and Load unpacked → open-dreamflow/extensions/chrome-bridge.

2. Open the extension popup to verify :3001 bridge health or click Capture Active Tab Media, or use the floating Send to Open-DreamFlow pill injected on Dreamina / Google Flow pages.

3. Events stream live over ws://localhost:3001/bridge (or click Test Bridge Capture inside the Browser Credit Bridge tab to spawn a live 35mm frame onto the canvas immediately).

License

Released under the MIT License.