I’m sure many of you have seen AI stuff popping up left and right these days. Given we have a slightly more niche interest, I was curious to see how far I could push a generator into spiting out crucifixion images. Looking online, I quickly ran into a couple issues: NSFW blockers, paywalls, and the fact that the image generators are really not well trained on crucifixion, so most of my images came out looking like old school paintings of Jesus.

Some of these, namely the paywalls and the content filters, can be avoided by running a local instance of Stable Diffusion, which I found relatively easy to get set up if you're willing to spend a few minutes on it and have a capable PC. From there, you can do a few things to get more realistic images, such as downloading checkpoints from civitai to get more realistic body generation. The biggest improvement by far for me, though, has been a tool called ControlNet. Now I'm far from an expert at using it, but it has allowed me to upload images and basically just have the AI redo a pre-existing image to be more in line with whatever you're after.

Now this still has a few limitations, namely that you are still stuck mostly with the same pose from the image and the AI still has issues. BUT it does give you a really good jumping off point!

I used this image as my starting image. I threw that in with the prompt:

A blonde woman, naked, outstretched arms, blonde hair, wearing a necklace, wrist tied, brick background, dim lighting, dark

And got this image. Honestly, this is one of the better first attempts I've seen it give me, but you'll still notice a few issues. Some things look weird, the cross in the background is all kinds of messed up, her eye is a bit odd, etc. Usually you'll also see it really struggle with noses....From here, you can either continue generating new outputs till you get one you're more happy with, or you can move on to the inpainting feature, where you can touch up the images spot by spot and have it re-generate only small parts of the whole. After several hours of messing around and trying over and over and over, I got a few I'm pretty happy with! See those below.

Image One!

Image Two!

Image Three!

Even these aren't perfect, but I thought it was interesting to see how much different you could make a couple images stemming from the same photo, and then a bit of something different as well! I'm far from an expert and hope to keep learning some tricks to further improve quality as time goes on, happy to try and answer questions anyone may have or provide anything unique if people want and I can figure out how to make it work as I have free time.