deepankar-sah/omniflow-studio

OmniFlow Studio Agent Skill

This skill grants AI coding and orchestration agents direct control over OmniFlow Studio — an autonomous bridge connecting local AI agents to high-fidelity generative media canvases in Google Chrome.

When to Use This Skill

Use this skill whenever:

1. The user asks to generate cinematic AI videos, B-roll footage, concept art, or storyboards.
2. Building or directing an autonomous multi-shot short film or ad creative.
3. Generating character reference sheets (turnaround consistency) across multiple video scenes.
4. Automating batch image/video generation queues without paying per-minute API fees.
5. Inquiring about current OmniFlow bridge status, generation quotas, or active queue progress.

---

🚀 Quick Execution Protocol for Agents

Before dispatching media jobs, every agent must follow this 3-step verification loop:

Step 1: Health & Bridge Check

Verify the local OmniFlow bridge daemon is running on http://127.0.0.1:8792:

node scripts/check-bridge.js
# Or curl directly:
curl -s http://127.0.0.1:8792/status

• If Offline: Instruct the user or run: omniflow bridge start (or start the bridge background task).
• If Online: Check extensionConnected is true and tabVisible is true (means Chrome has Google Flow open).

Step 2: Submit a Generation Job

Agents can submit via REST API or the provided helper scripts:

# Helper script (easiest):
node scripts/submit-shot.js --mode flow-text-to-video --prompt "Cinematic drone flyover neon cyberpunk Tokyo rain 8k" --aspectRatio 16:9 --duration 6s

# Or direct HTTP POST:
curl -s -X POST http://127.0.0.1:8792/jobs \
  -H "Content-Type: application/json" \
  -d '{
    "mode": "flow-text-to-video",
    "prompt": "Cinematic drone flyover neon cyberpunk Tokyo rain 8k",
    "params": {
      "aspectRatio": "16:9",
      "duration": "6s",
      "videoModel": "veo-3.1-fast",
      "downloadSubfolder": "MyProject"
    }
  }'

Step 3: Monitor & Retrieve Downloaded Media

Poll until the job status changes to completed:

node scripts/list-jobs.js
# Or:
curl -s http://127.0.0.1:8792/jobs

When status === "completed", retrieve the local file path from the files array (e.g. Downloads/MyProject/...mp4).

---

🛠️ Supported Generation Modes

Mode ID	Description	Key Parameters
flow-text-to-video	Generates motion video from a text description	prompt, aspectRatio (16:9, 9:16, 1:1), duration (4s, 6s, 8s)
flow-text-to-image	Generates photorealistic or stylized still images	prompt, aspectRatio, imageModel (nano-banana-pro)
flow-frame-to-video	Extends a still image or keyframe into continuous video	prompt, duration, reference asset token
flow-create-character	Generates 4-angle turnaround sheet for facial consistency	characterName, visualDescription, voicePreset

---

🎛️ MCP Tool Invocations (If MCP Server is Active)

If OmniFlow's MCP server is connected in your agent environment, call these native tools directly:

• omniflow_status: Check bridge and browser connection status.
• omniflow_generate_shot: Queue a single shot with camera movement (drone-wide, close-up, tracking).
• omniflow_produce_movie: Pass a complete screenplay array of ordered shots to produce an entire short film sequentially.
• omniflow_create_character: Lock a character tag with front & profile turnaround consistency.
• omniflow_list_jobs: List active, queued, and finished generation jobs with disk download paths.

---

💡 Best Practices for Cinematic Prompts

For photorealistic video outputs with Veo 3.1:

1. Specify Camera Motion: E.g. [Camera: drone-wide], [Camera: slow push-in], [Camera: tracking shot].
2. Specify Lighting & Atmospherics: E.g. volumetric rim lighting, golden hour anamorphic haze, 35mm film grain.
3. Avoid Jargon: Google Flow responds best to clear visual narrative descriptions rather than software tags.

See references/api-reference.md and examples/cinematic-short-film.json for ready-to-run multi-scene templates.