Today I downloaded Stable Diffusion which is an open text-to-image technology from Stability.ai that can render images on your own computer. You need a heavy duty graphics card but it’s free to use. The technology is open and many other start-ups are incorporating it into their systems including Midjourney which I have been very impressed with.
It’s early days and some of the installation methods are technically complicated (“step one: install Python programing language…” etc.) but I found a user interface project – Stable Diffusion UI – that has a “1 click” installer for Windows. It was a little more than one click, but I managed and now I’m rendering away testing it out.
My initial impression is that it’s not as good as Midjourney, not as artistic or flexible. The resolution is lower, 1024 x 1024 is the largest available in the drop down menu, at least in the GUI that I installed. I’ve also discovered that it has a tendency to create double headed characters if you use any resolution over 512 x 512. Apparently the AI was trained at that resolution and if you try something larger it tries to fill the space by duplicating things. Like I said, it’s early days.
I’m still very impressed though. I’ve only tried copying and pasting some of my midjourney prompts which were of course optimized for a different system. I have yet to spend time figuring out all the controls and I’m still getting some interesting images. They render fast, within a minute or two on my system which has nVidia Titan RTX graphics cards.
More to come…