Image to Textured 3d Mesh and Simple Crowd - Houdini with Trellis2

I’ve been experimenting with merging AI and VFX workflows for nearly 3 years now, and the progress is nothing short of incredible. My focus has been primarily on the ComfyUI – Houdini ecosystem.

Thanks to the Houdini ComfyUI Bridge (developed by Rafael Drelich and Anatolii Iudanov), I’ve built a prototype to test the Trellis2 model—now fully integrated within Houdini.

The goal was to Generate a Textured 3D Mesh, based on a single image.

While automated segmentation is still a hurdle and the output isn't quite ready for "Hero" assets yet, the leap in quality over just the last six months is staggering. Excited to see where this path leads! Test Scene:

I had the pleasure of appearing on Rafael Drelich’s YouTube channel last Sunday for a brief interview. I shared some insights into the image-to-3D mesh workflow—feel free to take a look!

Content of the video

1. Introduction & Professional Background (00:00:00 – 00:08:00)

2. Adoption of AI & The ComfyUI Philosophy (00:08:00 – 00:14:40)

The Node-Based Mindset: Why ComfyUI appealed to a Houdini artist—it offers a procedural, node-based logic that allows for artistic control rather than "random" generation.

3. Early Experiments in ComfyUI (00:14:40 – 00:22:30)

4. Technical Deep Dive: 3D Generation with Trellis2 (00:22:30 – 00:46:00)

5. Rigging, Retargetin Mocap Animation & Crowds (00:46:00 – 00:54:00)

6. Rendering & Future Automation (00:54:00 – 01:01:00)

Karma Rendering: Shading the AI models and rendering them within the Solaris/USD context.

PDG Integration: Discussing how Procedural Dependency Graphs can be used to mass-produce character variations automatically.

7. Exploring Gaussian Splatting (01:01:00 – 01:51:41)