Updated the installer to my new standards. It’s now better in every way.
Fixed Reactor nodes being persistently missing.
InsightFace now installs correctly (no more missing reactor nodes).
Added Mac support with simple terminal steps.
👉 If you ran any earlier version, run this one too. It won’t redownload files you already have.
Previous Updates (May 24)
Added two models for Face Swap and Face Restore to installer
Fixed missing note nodes in the workflow
Various tweaks and fixes
Download the Mega Workflow json file (bottom of this post).
Drag and drop into ComfyUI
It will give a popup if you are missing nodes
Click Open Manager → Install missing custom nodes
My nodes will remain missing, the installer will handle that
If the workflow looks empty, hit period key on your numpad to frame all nodes into view
Download the Launch_Mega_SDXL_Installer.bat and mega_sdxl_installer.py at the bottom of this post. (If you are on mac or linux, see below)
Place it directly in your ComfyUI folder

Double-click Launch_Mega_SDXL_Installer.bat
Answer a few short questions
✅ Let it run.
If you ran the v1 installer, run this one too. It will install anything you don't have yet.
Updated the installer to my new standards. It’s now better in every way.
Fixed Reactor nodes being persistently missing.
InsightFace now installs correctly (no more missing reactor nodes).
Added Mac support with simple terminal steps.
👉 If you ran any earlier version, run this one too. It won’t redownload files you already have.
Previous Updates (May 24)
Added two models for Face Swap and Face Restore to installer
Fixed missing note nodes in the workflow
Various tweaks and fixes
Download the Mega Workflow json file (bottom of this post).
Drag and drop into ComfyUI
It will give a popup if you are missing nodes
Click Open Manager → Install missing custom nodes
My nodes will remain missing, the installer will handle that
If the workflow looks empty, hit period key on your numpad to frame all nodes into view
Download the Launch_Mega_SDXL_Installer.bat and mega_sdxl_installer.py at the bottom of this post. (If you are on mac or linux, see below)
Place it directly in your ComfyUI folder

Double-click Launch_Mega_SDXL_Installer.bat
Answer a few short questions
✅ Let it run.
If you ran the v1 installer, run this one too. It will install anything you don't have yet.
Download only mega_sdxl_installer.py and place it in your ComfyUI directory.
Then:
Open Terminal
Type cd (with a space), then drag and drop the ComfyUI folder into the Terminal window, and press Enter
Run the installer:
.venv/bin/python mega_sdxl_installer.py
✅ It’ll ask a few setup questions and handle the rest.
⚠️ If you get a permission error, run:
chmod +x mega_sdxl_installer.py
Then try again.
Face swap, face restore, and image blending groups are included (Mega version only)
Workflows now use the new v1.5 layout with no hidden nodes, exposed wiring, and cheat sheet notes
Inpainting has been improved
Various custom node fixes and improvements
Windows Defender might freak out. That’s normal for unsigned apps.
Just click "More info" → "Run anyway"
(You just saved me over $100 🫡)
That's what v1.5 tries to make easier. What's the point?
Learning ComfyUI gives you:
Full control over your workflow layout and style
The power to chain things together (e.g. text to image → adetailer → upscale)
Mix and match techniques that typically aren't
Actually understand how ComfyUI is working
A lot of people hate ComfyUI because they misunderstand it and give up. This workflow is built so that doesn’t happen to you and puts you a mile ahead of the ones who bailed. 🫡
These aren't in the installer, why?
The installer has my own model (Hyper3D) and another for realism from the v1 installer.
This time, I wanted to set you up to start grabbing your own models because that's a core part of using ComfyUI long-term.
Here are a couple worth trying:
Pony Diffusion v6 – Anime
CyberRealistic XL – Realism
💾 To install them manually:
Place the safetensors files into ComfyUI/models/checkpoints/
I recommend checking out Civitai as that is where you will find great models and loras.
Random side note: absolutely remove "score_9, score_8_up... etc) from prompts if you aren't using Hyper3D. They can heavily affect the resulting image if the model wasn't trained on them.
Got questions, cool results, or weird generations to share?
👉 Join the Maxed Out Discord
It's also a great spot to contact me directly for questions. 🫡
Max
2025-08-20 08:46:09 +0000 UTCTigon
2025-08-19 10:28:05 +0000 UTC