- Home
- Skills
- Agents & Orchestration
- Flow Film Director
More screenshots
Works with the AI Agents you already use
Flow Film Director
Direct cinematic videos in Google Flow with consistent faces, products, and physically grounded actions.
$4.99
Flow Film Director
Example session with this skill installed
Make a 15-second cinematic product ad for our luxury chronograph watch. I have 3 reference photos of the watch and a reference video of a person walking through a modern park. The watch face, dial details, and metallic finish must remain 100% identical across all cuts without AI morphing.
- Read your context and instructions
- Compiled the flow film director
🎬 Flow Film Director — Production Plan Initialized
Project Slug: film/luxury-chronograph-ad/
Pipeline Mode: Mode A (Replicate & Anchor) | Aspect Ratio: 9:16 (Vertical Reel)
Audio Plan: Mixed (Upbeat ambient soundtrack + sidechain ducked voiceover)
1. Forensic Reference Audit & Object Locking (object_check.py)
- Cleaned 3 product references (cropped borders, normalized contrast to 1024x1024).
- Extracted 420 SIFT geometric keypoints on watch bezel, crown, and subdials.
- Object Lock Created:
bible/watch_chronograph.lock.txt(Geometric match score: 0.88 PASS).
2. Scene Breakdown & Storyboard Architecture
| Scene | Duration | Framing & Action | Camera Movement | Identity / Object Gate |
|---|---|---|---|---|
| 01 | 5.0s | Close-up on subject's wrist adjusting the brushed steel watch strap | Slow push-in on bezel | object_check: 0.84 PASS |
| 02 | 5.0s | Medium tracking shot walking through sunny park path, wrist raised | Lateral tracking at real-time pace | motion_match: 0.79 PASS |
| 03 | 5.0s | Hero macro shot: golden-hour sunlight sweeps across sapphire glass | Locked-off macro tilt | object_check: 0.91 PASS |
3. Zero-Credit Keyframe Gating (Phase 4)
- Compiled prompts via
build_prompts.pytoprompts/stage1-start-keyframes.json. - Generated 3 start keyframes for 0 Flow credits.
- Pixel verification: Verified bezel geometry and hands position anchored directly in
first_framepixels.
4. Post-Production Timeline Assembly
- Voiceover recorded & aligned with -14 LUFS loudness standard.
- Beat-synced cuts at timestamps
00:05.00and00:10.00. - Output render scheduled via Remotion programmatic React engine.
Next Step: Type "Animate" to render the 3 verified scenes in Google Flow (Estimated cost: 30 credits).
Connects securely to your tools. The creator never sees your data.
About this skill
🎬 The Deterministic AI Film Crew for Google Flow
Stop wasting hours and burning hundreds of Google Flow credits on morphing faces, rubber limbs, and inconsistent scenes.
Flow Film Director turns Google Flow (flow-mcp) into a disciplined, automated film production studio for Claude Code, Google Antigravity, and AI coding agents. Using a deterministic 5-stage pipeline, it ensures the same faces, products, and locations appear in every single shot — with physically grounded actions and frame-accurate editing.
🛑 Why AI Video Fails (And How This Fixes It)
| The Common Nightmare | What Flow Film Director Does |
|---|---|
| Drifting Faces: Characters look like different cousins in every shot. | Pixel Anchor Gating: Generates from verified keyframes. Identity lives in the pixels, not prompt luck. |
| Rubber Limbs & Glitches: Flow morphs bodies during complex actions. | Atomic Action Prompts: Compiles strict storyboard recipes with motion constraints. |
| Credit Burn: 500+ credits wasted on failed takes. | 0-Credit Stills Gating: Locks the visual style using free Nano Banana stills before spending video credits. |
| Robotic Voices: Generic TTS audio that doesn't match lips. | Voicebox Studio: 100% free, local on-device voice cloning with automated sidechain audio ducking. |
| Tedious Manual Editing: Hours trimming clips in desktop editors. | Remotion React Pipeline: Programmatic timeline assembly, kinetic typography, and broadcast MP4 export. |
🚀 What’s Inside the Pro Studio Edition ($4.99)
1. 🧬 OpenCV YuNet + SFace Computer Vision Gating
- Automated numerical facial likeness score ($\ge 0.50$ cosine similarity) measured against reference photos before any clip is accepted into your film.
2. ⚡ 95% Flow Credit Savings
- Validates beginning and ending keyframes with 0-credit Nano Banana images first.
- A 60-second narrative film that typically consumes 800+ credits finishes for ~35 credits.
3. 🏃 MediaPipe Action & Motion Tracking
- Compares generated movement against reference video breakdowns. Ensures characters walk, sit, or gesture at real-world physical velocities ($\ge 0.70$ motion compliance).
4. 🎙️ Jamie Pine's Voicebox Local Studio Integration
- Built-in support for the trending open-source Voicebox engine (
127.0.0.1:17493). - Replaces Flow stock audio with your characters' authentic voices (WhatsApp voice notes or audio clips) with zero API costs, zero data exfiltration, and Perth watermarking.
5. ✂️ Programmatic Remotion Timeline Assembly
- Declarative React video editing: beat-synced cuts, automatic sidechain dialogue ducking (-14 dB), glassmorphic lower-thirds, and burned-in subtitles.
6. 🔁 Fast Review & Auto-Retake Loop
- If a clip exhibits motion anomalies or identity slip, the engine diagnoses the exact failure point, adjusts only the offending storyboard variable, and retakes the shot automatically.
🛠️ What You Get in the Package
- Autonomous Agent Persona: Ready-to-run
video-creatoragent for Claude Code & Antigravity. - Master CLI Tool:
flow-film-directorcommand-line suite for auditing, scaffolding, and verification. - Flow MCP Compatibility Patch: Enables parallel batch submission, human-paced pacing (20–40s), and "unusual activity" detection.
- Reference Galleries & Prompt Playbooks: Tested recipes for cinematic lighting, shot sizes, camera motion, and character bibles.
- Complete Remotion Studio Workspace: Pre-configured React video project with spring physics and captioning components.
💡 Compatible With
Works out-of-the-box with Claude Code, Google Antigravity, Cursor, and any agent supporting MCP tools and Python 3.10+.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires flow-mcp and Python 3.10+. Faces should be at least 15% of the frame height for optimal identity verification. Generating video takes uses Google Flow credits, while starting keyframe stills cost 0 credits.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 day ago
- Accounts
GitHub
Google AI
- Software
PythonFlow MCPFFmpeg
Node.jsvoicebox
Permissions
Allowed Hosts
File Scopes
Permissions are strictly used to run local face/pose verification scripts (OpenCV & MediaPipe) and save project storyboards, audio files, and assembled MP4 video takes inside the local film/ folder.