top of page

ComfyUI Exploration

I've been curious to see how Ai can improve workflows, below are four of my ComfyUI projects and personal experiments.

01

Previous Company Work

I was part of a couple R&D projects where we used ComfyUI to generate imagery/matte paintings and environment assets such as: dead trees, candles, candelabra, chairs, tables etc.

xyz.skybox.player.ovr-20260302-155605.jpg
xyz.skybox.player.ovr-20260302-155615.jpg
xyz.skybox.player.ovr-20260302-155333.jpg
xyz.skybox.player.ovr-20260302-155347.jpg

02

Blender ComfyUI Add-On

I was interested to see how people were using ComfyUI with Blender and decided to test out generating my own other-worldly galaxy/nebula HDRI maps.

(https://youtu.be/TP97X627sBg?si=oXYHgOQgj1cIOdNA)

The add on works seamlessly with ComfyUI, it listens for updates and automatically adjusts the scene.

03

Majority Texture 3D Print Paint Job

This experiment explored whether ComfyUI could generate base colour textures to speed up painting for 3D printing, limited to four colours that I could later swap out.
 

I first tested feeding ComfyUI a UV island to see how well it could read the island's borders and the shapes sculpted in the model within the islands. It could not, even with a Canny Edge ControlNet.

 

I then tried a colour‑coded UV island to label different parts of the fish. Neither approach produced accurate, good or seamless results. The fix for this method would require ML training.

My workaround was to use an LLM to generate prompts that resonated better with ComfyUI, describing my desired art style, koi patterns and shapes without referencing fish directly.

Since topology and UVs matter less in 3D printing than in VFX, I used Blender to camera‑project the fish and adjust the UVs to apply the pattern.

 

Geometry Nodes were then used to bake the colour information so the slicer could identify it.

Above: Fish in Bambu Slicer

Above: Fish with GenAi Texture in Blender

Screenshot 2026-03-01 194737.png

04 

More Image to 3D Model

I used a LLM to generate images of elegant ornate frames, antique wall lamps and portrait spotlights.

 

Using the Hunyuan model, I then generated 3D models from those images.

The purpose of this personal project was to demonstrate layout, lighting, and rendering skills in a new software. ComfyUI helped speed up the asset stage by generating the more detailed assets.

Screenshot 2026-03-01 200553.png
Screenshot 2026-03-01 201325.png
bottom of page