I haven’t created any Gen AI deepfakes of myself since I last used Stable Diffusion in 2022. This week, I tried out the latest imaging framework called Ace++, created by researchers from Alibaba. You just need to upload a good portrait photo of yourself and the deepfakes will be stunningly good and unsettling. Notice how the lighting is different in each scene, yet the deepfake Ians all look quite naturally-lit.
In 2022, I had to upload about 30 to 40 photos to train a deepfake dataset of myself, and it took several hours for my gaming PC to get the job done. Today, it just takes one photo and perhaps a few minutes of waiting. Mindboggling.
If you want to try Ace++, do a search for it, but you also need to learn how to use ComfyUI and its crazy node (noodle) interface. All the tutorials are on YouTube, so go knock yourself out.
In other news, Singapore residents were scammed out of a record $1.1 billion in 2024. And it’s going to be much worse when the deepfake apocalypse begins.