Skip to content

deepankar-sah/OmniFlowStudio-Skills

omniflow-studio

Verified

Autonomous AI media generation, cinematic multi-shot film production, and Google Flow automation bridge. Enables AI agents to queue text-to-video, text-to-image, character consistency, and autonomous self-healing workflows with zero API costs.

OmniFlow Studio Agent Skill (v1.2.0)

This skill empowers AI coding and orchestration agents to act as Autonomous Film Directors and Generative Media Producers through OmniFlow Studio β€” a high-performance local bridge and Chrome Extension that controls Google Flow and Veo 3.1 canvases directly inside Google Chrome with zero API fees.


🎬 When to Use This Skill

Activate this skill whenever the user or task involves:

  1. Generating High-Fidelity Videos or Images: Photorealistic video clips, cinematic b-roll, stylized stills, or concept art.
  2. Autonomous Multi-Shot Film Direction (Director Mode): Multi-scene narrative continuity, character face/clothing consistency, and automated last-frame handoff between shots.
  3. Studio Max (10x Turbo) or Pro (5x Speed) media pipeline execution.
  4. Autonomous Error Recovery & Self-Healing: Resilient generation recovery against transient Google Flow failures or "unusual activity" notices without human intervention.
  5. Checking Quotas & Bridge Status: Verifying local bridge health, active generation queue progress, or seat licensing.

πŸ€– The Agent-Centric Director Protocol (Zero UI Overhead)

OmniFlow operates under an Agent-Centric Architecture: there are no confusing browser UI buttons for the user to fiddle with. The AI agent manages the entire production cycle directly via native Model Context Protocol (MCP) tools or local HTTP endpoints.

πŸ’¬ Conversational User Alignment (Interview First)

Before dispatching scenes to Google Flow, the AI agent must conduct a brief, friendly interview with the user to establish the creative vision:

  1. Narrative & Story: What is the core storyline, theme, or concept?
  2. Visual Aesthetic: What lighting, color palette, and lens style are desired? (e.g. Anamorphic Cyberpunk, Golden Hour 35mm, Studio Ghibli Anime, Dark Nordic Noir).
  3. Aspect Ratio:
    • 16:9 (Cinematic Film, YouTube, Desktop)
    • 9:16 (Vertical Reels, TikTok, YouTube Shorts)
    • 1:1 (Square Social Media)
  4. Camera Language: What camera movements best tell the story? (drone-wide, slow tracking push-in, extreme close-up, orbit 360, low-angle tilt).
  5. Speed & Anti-Detection Profile: balanced (recommended default), stealth (safe human cadence), or turbo (10x Studio Max).

Crucial Rule: The agent must always announce its chosen parameters and storyboard plan to the user before dispatching render jobs!


🎬 The 5-Step Director Mode Workflow

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚ 1. Initialize Session ('flow_director_session')        β”‚
β”‚    Set Title, Genre, Visual Aesthetic, Aspect Ratio    β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                            β”‚
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚ 2. Character DNA Lock ('flow_director_character')      β”‚
β”‚    Render 4-angle turnaround sheet to anchor identity  β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                            β”‚
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚ 3. Register Storyboard Plan ('flow_director_plan')     β”‚
β”‚    Validate ordered scenes and camera cinematics       β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                            β”‚
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚ 4. Sequential Scene Generation                         β”‚
β”‚    ('flow_director_generate_scene')                    β”‚
β”‚    Auto-extracts Scene N-1 last frame for continuity   β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”¬β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜
                            β”‚
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β–Όβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”
β”‚ 5. Lossless Master Cut Stitch ('flow_director_stitch') β”‚
β”‚    Concatenates scene MP4s into final film in Downloadsβ”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

πŸ› οΈ Complete MCP Tools Catalog

Add OmniFlow's MCP server to your agent's config (mcpServers):

{
  "mcpServers": {
    "omniflow": {
      "command": "npx",
      "args": ["-y", "omniflow-studio", "mcp"]
    }
  }
}

1. Director Mode Tools

Tool NamePurposeKey Parameters
flow_director_sessionInitializes production session settingstitle, genre, visual_aesthetic, aspect_ratio (16:9, 9:16, 1:1), speed_profile (balanced, stealth, turbo), default_duration
flow_director_planRegisters an ordered storyboard of shotsscenes: Array of { scene_number, prompt, camera, duration, continuity_from_previous, character }
flow_director_generate_sceneGenerates a specific planned scene on Google Flowscene_number, video_model (veo-3.1-fast, veo-3.1-quality, veo-2)
flow_director_characterLocks character facial & wardrobe consistencyname, visual_description, voice_preset
flow_director_stitchLosslessly joins scene MP4s into a master cutscene_files (optional), output_filename
flow_director_statusReturns production state, active jobs, and file paths(none)

2. Studio Control & Error Recovery Tools

Tool NamePurposeKey Parameters
flow_retryAutonomous in-page self-healing for failed tiles(none) β€” Automatically finds unhandled failed tiles
flow_get_settingsInspects Google Flow settings across 7 categories(none)
flow_configure_settingsConfigures mode, quality, duration, aspect ratio, modelmode, aspect_ratio, quality, model, duration, outputs_per_prompt
flow_inspect_stateReturns live DOM state: title, characters, media(none)
flow_upload_characterUploads character reference image to Flow libraryname, file_path or base64_data, mime_type
flow_add_ingredientsAttaches reference image into Flow prompt barfile_path or base64_data, filename
flow_manage_projectCreates a new Google Flow project canvasaction: 'create-new'
flow_configure_delaysConfigures anti-detection typing speeds and delaysprofile: 'stealth' | 'balanced' | 'turbo'

3. Core Generation & Queue Tools

Tool NamePurposeKey Parameters
omniflow_statusChecks Bridge daemon and extension connectivity(none)
omniflow_list_jobsLists all historical and active media jobs(none)
omniflow_generate_shotQueues single standalone video or image shotprompt, mode, aspect_ratio, duration, camera_movement
omniflow_create_characterGenerates turnaround reference sheetname, visual_description, voice_preset
omniflow_produce_movieHigh-level screenplay-to-film pipelinetitle, shots, aspect_ratio

πŸ›‘οΈ Autonomous Error Recovery & Rate Limit Self-Healing

OmniFlow features an autonomous in-page error recovery system that monitors the Google Flow DOM for:

  • "Unusual activity" or rate limit warnings
  • "Generation failed" or "Something went wrong" tile cards

How the Recovery Engine Works:

  1. Strict 1-Retry Cap per Tile: To prevent infinite loops and runaway generation costs, each failed tile is retried at most 1 time.
  2. Dual-Layer Deduplication:
    • The failed button and card are tagged in the DOM with data-omniflow-retried="true".
    • The tile's spatial fingerprint is registered in an in-memory Set and WeakSet.
    • Query selectors use :not([data-omniflow-retried]) to guarantee handled cards are permanently ignored.
  3. Visual Lock: The retried button is dimmed (opacity: 0.4; pointer-events: none) so neither scripts nor accidental clicks can trigger duplicate runs.
  4. Circuit Breaker Cooldown: Enforces a 25-second minimum backoff between consecutive recovery cycles.
  5. Manual Trigger: Agents can also manually invoke recovery at any time using the flow_retry MCP tool or POST http://127.0.0.1:8792/flow/retry.

⚑ Quick Start for AI Agents

Step 1: Check System Health

Run npx omniflow-studio doctor or query omniflow_status:

npx omniflow-studio doctor

Ensure:

  • Bridge is listening on http://127.0.0.1:8792.
  • Extension is connected.
  • Google Flow tab (https://flow.google.com) is open and logged in Chrome.

Step 2: Start Bridge Daemon (if offline)

npx omniflow-studio bridge daemon

Step 3: Directing a Film via MCP

// 1. Start Session
await callTool('flow_director_session', {
  title: 'Neon Odyssey',
  genre: 'Cyberpunk Sci-Fi',
  visual_aesthetic: 'Anamorphic lens flares, volumetric rain, neon blue and amber rim lighting, 8k hyperreal',
  aspect_ratio: '16:9',
  speed_profile: 'balanced'
});

// 2. Register Storyboard Plan
await callTool('flow_director_plan', {
  scenes: [
    {
      scene_number: 1,
      prompt: 'A solitary rogue hacker in an iridescent trenchcoat stands on a rain-slicked skyscraper balcony looking over Neo-Shibuya',
      camera: 'drone-wide',
      duration: 6,
      continuity_from_previous: false
    },
    {
      scene_number: 2,
      prompt: 'The hacker steps forward, pulling up a glowing cyan holographic data gauntlet while sirens reflect in the puddles',
      camera: 'tracking push-in',
      duration: 6,
      continuity_from_previous: true
    }
  ]
});

// 3. Generate Scenes
await callTool('flow_director_generate_scene', { scene_number: 1 });
// Wait for Scene 1 completion, then generate Scene 2 with automatic frame chaining:
await callTool('flow_director_generate_scene', { scene_number: 2 });

// 4. Stitch Master Video
await callTool('flow_director_stitch', { output_filename: 'Neon_Odyssey_Master.mp4' });

πŸ“š Reference Links

reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why β€” repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" β€” or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself β€” your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall