- AI-Enhanced Creator
- Posts
- Building with Benjamin: Crafting a Custom Comfy Node from Scratch
Building with Benjamin: Crafting a Custom Comfy Node from Scratch
Full Breakdown Inside + New Cohort Details 🌞
Get Ready to Create Fully Custom AI Workflows
Hello Friend,
Adam here, sitting clad in a beanie and tracksuit because it’s even greyer, rainier, and windier than when last I wrote to you. The saving grace? It’s the kind of weather that makes you want to stay inside and create something just for the joy of it.
I couldn’t think of a better company to work with than Lighthouse AI Academy, purely for that reason!
In our last newsletter, we explored a Hollywood-level production pipeline with our mentor, Lovis Odin.
This time, we’re sailing straight into the code.
In a recent live session, Lighthouse mentor Benjamin Bertram showed our students how to build a custom ComfyUI node from scratch.
Not tweak. Not tinker. Build.
If you’ve ever hit a wall with ComfyUI and thought, “If only this node did exactly what I need,” -- this one’s for you.
Before we get to it, a few updates from the team 👇🌞
New: Lighthouse AI Academy Now on YouTube 🌐
We’ve officially launched our YouTube channel — a place for public tutorials, session snippets, and creative showcases.
👉 Subscribe here
New Cohorts Are Open For:
Advanced ComfyUI (Cohort 2) — Starts 21 August
→ For building workflows and deploying real-world tools.
→ Ideal for creative technologists, VFX leads, devs, and directors.
→ Apply to join nowCreative Leaders (Cohort 3)— Starts 9 September
→ For mastering AI’s creative, ethical, and strategic dimensions.
→ Ideal for directors, producers, team leads, and educators.
→Apply to join now
Inside the Node: Benjamin’s Custom ComfyUI Workflow
Here’s how our mentor, Benjamin, walked through building a node from scratch — no guesswork, just a real workflow students can replicate.
1. → Set up the workspace
Students opened their node folder in VS Code. Two files were created:
• oyden_denoiser.py — where the logic lives
• init.py — to register the node in ComfyUI

2. → Define what the node should do
Benjamin had students clarify:
• What inputs and outputs will this node handle?
• What’s the function -- in this case, image denoising?
• What’s already out there, and what are we adding?
• Does it run on CPU or GPU, and what libraries are needed?

3. → Write the prompt
He walked through how to engineer a precise, scoped prompt for coding agents like Claude, Rootcode, Groq, and ChatGPT. Key instructions included:
• Clean function names
• Adjustable denoising strength via slider
• Output: clean image
• Use Intel’s open-image-denoise
• Include error handling, README, requirements.txt, and code comments

4. → Live coding with agents
Students followed along as multiple LLMs generated the node logic. Rootcode handled the full build from start to finish inside VS Code.
The final version worked across CPU and GPU, had functional sliders, and was fully documented.

5. → Wrap-up & testing
The node was ready to use immediately — no extra config.
It also worked privately, meaning it didn’t need to be published or registered publicly to function inside ComfyUI.

For many of our students, the big takeaway was simple, but powerful:
You don’t have to wait for someone else to build the tools you need.
You can do it yourself, and now you know how. 🌞
Homework Time: What Node Would You Build?
Ask yourself:
What part of my ComfyUI workflow feels out of my control?
If I could write just one node, what would it do?
What tool do I wish existed?
Sketch it. Name it.
And if you want to build it?
That’s exactly what we’re here for.
Follow the Light & Keep Learning
This is about more than AI; it’s about ownership.
Inside Advanced ComfyUI, we teach workflows like Benjamin’s from the ground up, helping you turn them into usable tools.
Inside AI for Creative Leaders, we teach you how to scale those ideas across teams, campaigns, and entire production pipelines.
If you’re ready to build instead of borrow, we’re ready to show you how. 💛
--------------------------------------------------------------------------------
That’s it for now: Thanks for reading and building this new era with us.
Step by step, node by node — and away we go!
Keep creating and always remember to have fun.
— Adam & the Lighthouse AI Academy Team ☀️