← sources

[Open Source] ComfyUI Motion Previs Studio — 3D Camera Blocking, Character Animation, OpenPose/Depth Exports & Direct ComfyUI Rendering

source: reddit (reddit:comfyui)published 2026-10-30 20:52ingested 2026-10-31 02:39

retained text

Hey everyone!

I've been working on an open-source project called **ComfyUI Motion Previs Studio**, and I'd love to get some feedback from people who are seriously into ComfyUI, AI filmmaking, and advanced video generation workflows.

The idea came from a problem I kept running into:

**We have increasingly powerful video generation models, but precise control over camera movement, character blocking, spatial composition, and action choreography is still challenging.**

Instead of relying entirely on long prompts and repeated generations, what if we could actually **direct the scene in 3D before sending it to a video model?**

That's what I'm trying to build.

# What the project does

It's a local-first, interactive 3D previsualization studio designed to work alongside ComfyUI.

The current implementation includes:

* **Dual 3D viewports:** A free-orbit scene editor alongside a live cinematic camera viewfinder.
* **Camera control:** Dolly, orbit, tracking, crane, handheld, whip pan, and customizable camera trajectories.
* **Character animation:** Procedural walking, running, jumping, combat actions, gestures, and interactive FK/IK bone posing.
* **Scene building:** 3D environments, props, lighting rigs, and reference-image-based scene configuration.
* **Cinematic framing:** Adjustable focal lengths, depth of field, aspect ratios, and shot composition.
* **24fps timeline:** Camera and character keyframes for planning shots before generation.
* **Control passes:** Export MP4 previsualization, OpenPose skeletons, depth maps, alpha masks, and camera transformation matrices.
* **AI integration:** Optional support for local Ollama/LM Studio models or cloud models for natural-language scene direction and reference analysis.

# ComfyUI integration

This is the part I'm particularly interested in developing further.

The studio is designed to export reference and control assets for workflows involving LTX-2.5, MiniMax H3, Seedance, and other video models that support reference inputs.

It also includes a direct local ComfyUI API bridge for a specific **MiniMax H3 Ref2VA workflow**, allowing the studio to submit its previs video and reference images to ComfyUI and display the generated result.

The goal is to move toward a workflow like:

**3D Scene → Camera & Character Animation → Control Passes → ComfyUI → AI Video**

Rather than generating blindly and trying to correct everything afterward.

# Why I'm sharing this here

I'm not presenting this as a finished production-ready alternative to established 3D software.

It's an evolving open-source project, and I think the ComfyUI community is exactly where the more interesting technical discussions can happen.

I'd especially appreciate feedback on:

1. **ControlNet integration:** Which pose, depth, or camera-control workflows would be most valuable to support directly?
2. **Video models:** Should I prioritize deeper LTX integration, Wan workflows, or improving the existing MiniMax H3 bridge?
3. **Workflow interoperability:** Would you prefer importing/exporting standard ComfyUI workflow JSONs, or a more tightly integrated API-driven approach?
4. **Production usability:** What's missing from current AI video workflows when it comes to reliable camera motion and character choreography?

The project is **open source under the MIT license**.

**GitHub:** [https://github.com/navarisun1982/comfyui-motion-previs-studio](https://github.com/navarisun1982/comfyui-motion-previs-studio)

The repository includes screenshots, a demo video, installation instructions, and the source code.

I'd genuinely appreciate technical feedback, ideas, bug reports, or contributions from anyone working on similar problems.

**My bigger goal is to make AI filmmaking less about generating random results until something works, and more about actually directing the shot.**

Thanks for checking it out!

document fetched 2026-10-31 00:44 · canonical https://reddit.com/r/comfyui/comments/1xjr2ia/open_source_comfyui_motion_previs_studio_3d