ComfyUI-scene-camera-action
Playable 3D previz for AI video. Stage environments, act with keyboard controls (WASD), cut dynamic cameras, and feed exact motion into generative video models.
Nodes (8)
ComfyUI-scene-camera-action
Interactive 3D scene staging, actor acting, and camera directing suite for ComfyUI.
Created by @arturitu / Unboring.net.
Stop prompting movement into a black box. ComfyUI-scene-camera-action turns AI video generation into an interactive, playable 3D previz studio:
- Staging: Build navigable 3D blockout sets with instant colliders — manually or via AI prompts (
SKILL.md). - Acting: Playable
WASDcontrols & multi-track actor chaining with live ghost replays (no keyframing!). - Directing: Live camera cuts with anti-collision Spring Arm, outputting crisp 720p HD reference video (
VIDEO) for models like Seedance 2.5 / 2.0 Fast or MiniMax H3 Reference-to-Video.

⚡ Quick Installation
Option 1: ComfyUI Manager (Recommended - 1-Click Install)
- Open ComfyUI.
- Open Manager ➔ Custom Nodes Manager.
- Search for
Scene Camera Action(orcomfyui-scene-camera-action). - Click Install, then restart ComfyUI.
Option 2: Comfy Registry CLI (comfy-cli)
comfy node install comfyui-scene-camera-action
Option 3: Manual Git Clone
Open your terminal in your ComfyUI directory:
cd ComfyUI/custom_nodes
git clone https://github.com/arturitu/ComfyUI-scene-camera-action.git
pip install -r ComfyUI-scene-camera-action/requirements.txt
Windows Portable Users: Run this from your
ComfyUI_windows_portableroot folder:python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-scene-camera-action\requirements.txt
Restart ComfyUI after installation. The nodes will appear in the scene-camera-action category.
🎯 Ready-to-Use Example Workflows (Drag & Drop)
We provide pre-built workflows in the examples/ folder so you can start directing immediately without any setup:
| Workflow | Description | Download |
| :--- | :--- | :--- |
| 🟢 Basic 3D Previz (Recommended) | 0 AI models required! Pure 3D Previz pipeline (Staging ➔ Acting ➔ Directing ➔ Save Video) with race_track.json preset & Car actor ready to drive, including AI reference video integration. | basic_previz_workflow.json |
| 👥 Multi-Actor Chaining | industrial_ruin.json preset with Human and Quadruped actors on distinct spawn points, synchronized ghost replays, and AI reference video integration. | multi_actor_workflow.json |
💡 How to load a workflow:
- Method A: Download any
.jsonfrom above and drag & drop it directly into your ComfyUI browser window. - Method B: Click Load in ComfyUI and select the file from
ComfyUI/custom_nodes/ComfyUI-scene-camera-action/examples/.
🎬 3-Step Creator Guide (No Programming Required)
https://github.com/user-attachments/assets/dd5c0d21-573a-467f-9d4b-1218e2c90cac
1. Staging (StagingNode) — Build your 3D Environment
- Load Presets: Pick built-in 3D environments from the preset dropdown (
courthouse_square,gas_station,liberty_beach,industrial_ruin,warehouse, etc.). - Interactive Viewport: Orbit (Left Click + Drag), Pan (Right Click + Drag), and Zoom (Scroll).
- Transform & Compose: Add primitives, move, rotate, and scale assets. Use
Shift + Clickfor multi-selection and Grouping (Group/Ungroup). - Output: Connects
Stage Datadownstream into the Acting node.
2. Acting (ActingNode) — Drive & Record Performances
- 3 Actor Archetypes:
- 🚶 Human: Rigged 3D character with automatic walking, running, and idle locomotion.
- 🚗 Car: Vehicle physics with dynamic steering, acceleration, and drift.
- 🐕 Quadruped: 4-legged animal archetype with full locomotion animations.
- Game-Like WASD Controls: Move around the stage in real time using
W,A,S,D. - Practice Rehearsal Mode: Test movement loops freely before recording.
- Multi-Actor Chaining: Connect multiple Acting nodes in series. Sequence independent actors while watching synchronized live ghost replays of previous takes!
- Output: Combines stage geometry and actor trajectories into
Acting Data.
3. Directing (DirectingNode) — Compose Camera Cuts & Capture HD Video
- Smart Camera Rigs: Switch between Third-Person (TPV), First-Person (FPV), Tracking Side, and Auto-Framing Master Wide.
- Anti-Collision Spring Arm: Real-time obstacle avoidance prevents camera clipping against walls and geometry.
- Visual Cut Timeline: Add and edit camera cuts on the fly along the playback timeline.
- Direct-to-Generative Output: Outputs crisp 720p HD
Captured Video(VIDEO) ready to plug into reference-to-video AI models!
🤖 Transform Prompts into 3D Scenes (SKILL.md)
Generate proportioned 3D previz scenes from natural language prompts or reference images using the included stage-builder skill with any AI coding agent (Antigravity, Claude Code, Codex, Cursor, Pi, etc.).

The skills/stage-builder/SKILL.md instruction set equips AI coding assistants with 3D spatial reasoning to output clean 3D SceneState JSON preset files saved to presets/.
💡 FAQ & Troubleshooting
<details> <summary><b>Q: Directing node gives an error saying "Directing canvas is disabled. Connect an Acting node and record motion first."</b></summary> <b>A:</b> Directing requires actor motion to frame the cameras. Make sure you select the Acting node, click <b>Record</b>, and move your actor using <code>WASD</code> before queuing the Directing node. </details> <details> <summary><b>Q: Does this require a powerful GPU to run the 3D viewport?</b></summary> <b>A:</b> No! The 3D viewport runs on standard WebGL in your browser and features <i>Demand-Based GPU Rendering</i>, meaning it automatically pauses rendering during graph navigation for lightweight performance. </details> <details> <summary><b>Q: How do I feed the video into Seedance / MiniMax?</b></summary> <b>A:</b> The Directing node outputs standard ComfyUI <code>VIDEO</code>. Simply connect the <b>Captured Video</b> wire into the video input of your chosen reference-to-video model (e.g., Seedance 2.5 / 2.0 Fast or MiniMax H3 Reference-to-Video). </details>🚀 Release History
🚀 Shipped - v0.5.0: Quadruped Actor (Current Version)
- Rigged Quadruped Actor: New 4-legged animal archetype (
QuadrupedActor) with full locomotion animations (idle, walk, run, sit/crouch) built onSkinnedActor. - Configurable Actor Scaling: Adjust actor sizes dynamically with automatic camera framing and speed scaling.
- Practice Rehearsal Mode: Loop actor movement in real time to test trajectories before recording.
- Demand-Based GPU Rendering: Pauses WebGL rendering during graph navigation for lightweight performance.
- Stage Distance Fading & UI Upgrades: Smooth edge fading, modern Lucide icons, and optimized node payloads.
https://github.com/user-attachments/assets/d8abce40-8af8-486d-b8ac-fbcf2a661968
📦 Shipped - v0.4.0: Multiple Actors
- Multi-Actor Chaining: Sequence and record multiple independent actors (humanoids and vehicles) on the same stage with synchronized playback.
- Camera Spring Arm (Anti-Collision): Real-time obstacle avoidance prevents camera clipping against walls and geometry during dynamic camera moves.
- Actor Color Customization: Dynamic mesh coloring to visually distinguish different actors across the 3D viewport and node timeline.
- Universal Stage Builder: Universal 3D geometric engine with 5 core primitives and new presets (
courthouse_square,gas_station). - Vue.js Architecture: Modernized modular reactive UI components (
StagingWidget,ActingWidget,DirectingWidget).
https://github.com/user-attachments/assets/2d03f9fe-42a2-4472-9543-373252dcf670
📦 Shipped - v0.3.0: Animated Humanoid
- Animated 3D Humanoid Model: Custom 3D human model with bone structure and armature animations.
- Configurable Spawn Points: UI selection for actor starting locations across Staging and Acting nodes.
- 3D Scene Presets: Built-in previz tracks (
liberty_beach,industrial_ruin,warehouse,test-collider).
https://github.com/user-attachments/assets/590b10b8-fe3f-4f4b-b1be-99f3fc646010
👉 See it in action on LinkedIn
📦 Shipped - v0.2.0: Car Actor
- Natural Language 3D Scene Builder (
SKILL.md): Transform prompts or reference images into 3D stage compositions using any AI coding assistant. - Multi-Archetype Actor Physics Engine: Support for Humanoid and Vehicle (
CarActor) movement physics. - Hierarchical Scene Management: Grouping (
Group/Ungroup), multi-selection, and transform snapping.
https://github.com/user-attachments/assets/fc99bde7-7900-4fb1-9181-6c82dc451373
👉 See it in action on LinkedIn
📦 Shipped - v0.1.0: Starting Point
- Initial PoC Custom Nodes: Proof-of-concept custom nodes (
Staging,Acting,Directing). - Basic 3D Prev-viz: Asset placement, single capsule actor control, and multi-camera timeline cuts.
https://github.com/user-attachments/assets/fde67050-c33d-406c-87e5-64b9d0544381
👉 See it in action on LinkedIn
📄 License
MIT License. Built with ❤️ by @arturitu / Unboring.net.