High resolution real-time interactive video creator and dream explorer.
Prior to Oct 2023 dreams of realtime AI videos were just a future promise. Just after I retired as a performance architect from Microsoft in early 2022 I learned of Stable Diffusion and I was hooked. I spent very little time on the art side and 99% on making SD faster. In Oct 2023 Simian Luo announced LCM and I saw that the time had arrived. Within hours of LCM dropping I had what might have been the first real-time video generations running on my 4090 and I posted examples to reddit and elsewhere.
At 512x512 it was just for fun to do RT deepfakes where I looked like Trump/Biden or Emma Watson/Tom Cruise on camera. Over time new model compilers or other optimizations techniques appeared and I leveraged them into complete solutions to speed things. As my frame rates kept improving to above 50fps or 100fps with low quality sdxs(just a TECH demo) I realized I might be able to do much larger formats and with SDXL quality. I got 800x800 sdxl to 33fps, 1024x1024 sdxl to 22fps, and 1280x1024 to 17fps. At this point I knew I needed to do something more than just a twitter post like:
or my 6 months old reddit posts. But I'll never stop looking for ways to keep just pushing the bleeding edge of performance.
This is the beginning of a project to achieve larger goals.
Open source is down the road given I don't have the resources of a $10 billion company like Midjourney which just announced the long term goal of doing what I've been doing for months. My one-man show will go into stealth mode till I have a more complete product. I have two very talented guys that have volunteered to assist me.
To know the direction I'm headed watch the twitter video. Maximize the volume as my voice was a bit low.