AI – LivePortrait and Stable Diffusion, deepfake and imagination

My life and my work always been about creating solutions, solving problems and improving things by automating. We all know ChatGPT. When it became accessible to everyone, it scared people in the tech industry and as a self learner, I was (?) scared for one thing about AI; someone will make money off it and it won’t be me. I just said it at the beginning; when there is a problem, there is a solution. Well, let’s experiment AI.

The problem when you get into this, it is very technical and people lose interest quicky. But it can be simplified. This is where people make money; by building online solutions like Mid-Journey because geeks lose people in a conversation after 30 seconds even when you watch their YouTube tutorials. I am not the normal geek but I am damn good at self learning and let me tell you; AI is fabulous and is actually changing how I see its usage and you should to because it’s there to serve you more than you think. From finding de-aging actors, to write code to find new cure for diseases.

I’ll be quick on this topic and as simple as I can. What AI stands for ; Artificial Intelligence. How does it work? One or more computers are compiling massive amount of data (picture, text, voice, music, video…) and companies are creating algorithms (program) to train another set of computers to recognize what has been ingested. We call them models! The data comes from the Internet mostly (Youtube, Google, Facebook, but also use by your insurance company, credit companies and more to predict human behavior to tailor your experience and sell you that you fits your need).

Once the models are created, they can be used by computer(s) with AI abilities to translate instructions (that we call prompts). Most recent computers have AI chips (even phones now) and the most powerful “at home” for consumers are (expensive) graphic cards (GPU) that NVidia (the most valuable company at the moment) are selling and interpret/translate your prompts from these models (big files).

Anyone is able to train and create models (LORA). Example, you could take four pictures of you in four different seasons and mix them with already trained models like Stable Diffusion. With specific prompts (and the more you are precise, the more detailed will be the output) you could example type ; a summer day, a 30 y.o. Japanese lady, taking care of a dragon, field of barley. And you will get a result like this image below. Nice wallpaper hey?

This is where your imagination starts. Stable Diffusion is at the core of the existing models so far. It is beyond limits when you mixed it with the AI community like civitai.com (where you can download tons of LORA with specially crafted models).

Did you ever imagine Tom Cruise as Superman? Here you go;

Recently another Open Source application (LivePortrait) has been released where you can make a picture of someone’s face to talk based on a video of someone else talking with their gesture. It’s early, but the results are pretty interesting.

Let think about someone that never seen Joe Biden in his life, but hear only his voice for the first time from far. What would be the characters that represent him the most. I don’t want to do politics, but this one was really funny to create. The Old Farts from The Muppets came to my mind directly. Enjoy 🙂