With the internet in the hands of a few holders, I recently decided to create my own Metahuman avatar. The idea is to start using her instead of my own face on social media, which is not by nature decentralized (such as Mastadon, Lens Protocol, BlueSky, Farcaster, or similar). It will disable the creation of more content that potentially makes my face used for further AI training or other things that I didn't agree on. Those are my motivators, but with this specific project, I wanted to push the boundaries and see what is possible. So let's get started!
ComfyUI is the primary tool required for this workflow stack. If you have heard of the software before but are unsure where to start, let me try to make it easy by giving you some directions. First of all, let's answer the questions about why you should use Comfy over other tools:
ComfyUI is an open-source technology that will always be available to the community. Because of its open-source nature, it has an incredibly large community that builds on the original tool (hence the many freely available workflows and custom nodes).
Creators already suffer from subscription fatigue. Instead of subscribing to four tools, use one and generate images on your device directly without additional cost.
IP (Intellectual Property) rights take a whole different form when you utilize your device compared to centralized platforms. While generating images with tools such as Midjourney does not enable you to hold copyright, using a tool that requires human creative choices, such as ComfyUI, copyright is potentially in your hands. This is something to consider if these tools are used in a professional capacity.
There are more benefits, but I feel the above ones are already very convincing. That said, Comfy recently made the installation super easy. The only thing you need to do is download the setup file from their official page here.
To turn NOVA (my personal avatar) into a fashion icon, I had to go "undercover" to get some research work done before I became sure that some workflows work better than others. Here are my results so you don't have to do the same job all over again. Remember that all ComfyUI-generated images contain the workflows that were used to generate them. Here are the different steps I took to get my final result:
PART 1
I started by generating a color palette to my liking. You can use whatever makes you happy, but to me, colors are an incredibly defining element of my design work. For this, I created a workflow from scratch in ComfyUI that generates color palettes in the form of makeup palettes for me. I use this step to add a more personal touch to the design we will generate in PART 2. Here is the color palette I ended up using:
Most importantly, enjoy the process and have fun. Don't get angry over installations of custom nodes; instead, get a nice beverage or some sweets and get back to it with a clear head.
I am not going to lie. Seeing NOVA's face gracing the following few pages made me feel proud, but it also made me crucially aware that these powers are so very easy to abuse. Please make sure to utilize these tools consciously. They are beautiful and precious, but they shouldn't be used to disrespect, infringe or financially harm others!
 |
If you enjoy my writing, please consider supporting me. There are several ways of doing so, and not all of them cost money:
Subscribe to this blog. Thank you if you have already subscribed.
Get my free or paid membership tier powered by Unlock Protocol (it will unlock parts of my WIP website).
Leave a tip using the recently implemented system here on Paragraph by clicking the "Support" button at the top of the article and leaving a tip. Thank you if you have already done so.
PART 2
This step is all about defining your overall silhouette and composition of your editorial design. I was inspired by Mermaid Core but wanted to bring it to a level that feels sleek and elevated without embellishments. I used the image from PART 1 to influence composition, color, and shape. This is one of the results I got:
PART 3
In this next step, we are working on one of the more challenging elements of this workflow stack: swapping out the original face for my personal avatar, NOVA. I couldn't have figured this out myself, which is why communities around open-source technologies are amazing and beyond helpful. Thank you to Sebastian Kamph for all the fantastic work you do. I highly recommend subscribing to his YouTube and potentially to his Patreon page! This is the tutorial I used to figure out how to place NOVA's face on top of the original rendering from PART 2:
And here is my interim result:
PART 4
Looking at my interim result from PART 3 might already tell you what we should fix last. Right, the hands. Another workflow where there are a lot of recommendations, but I couldn't get all of them to work for me. The one I stuck with was this one by Motor_Perspective837. Be prepared, this workflow requires a bunch of custom nodes and even additional models. It reads the existing hand(s) and segments them into a depth mask. With that information plus specific prompts, it will attempt to generate new hands and place them. For one of my final results, I was faced with a challenge I couldn't have overcome without utilizing some custom editing and painting in Photoshop. Here is my final result:
 |
 |  |  |
Share Dialog
Support dialog
