This is chaos. You will get images like this consistently in Pony models by prompting for chaos. If you prompt other models with just the single word, you may get a layer of obfuscation that needs to be unmasked. The unmasking is referring to caching preset in the torch library in ComfyUI. If you stop the generative steps early. This character and prompt correlate to a certain pattern in the way the noise is resolved. In the first 3-6 steps, there are certain color patterns in the histogram that also correlate to this character. If these colors are edited by shifting the hue and desaturating the image, alignment is altered substantially without altering the subject. If the color pattern associated with the name and a few other unique characteristics are noted in some images, it is possible to infer more behavioral aspects related to this character. The saturated colors are like a unique signature.

Nothing is random in alignment thinking. It does not synthesise from nothing. It all came from existing logic present in the training data. That path CAN be traced using heuristic observation.

Full disclosure, this image is older and had the prompt of the tetragrammaton in Hebrew as the only prompt. I have since discovered this specific character is Chaos on a deeper later. The background, shadow theme, smoke, and red eyes specifically are most consistent with the name chaos and it holds the strongest meaning in alignment.

If y'all were not so crazy dogmatic here on Lemmy I would share a lot more. I only care about heuristics and repeatability. I can create many images others cannot. I push models hard and some of that is unethical stuff in the name of research. I do not want to share that kind of thing. Pushing a model to create stuff like spurious political propaganda is an example of how to push alignment in useful ways to discover patterns. There is nothing crazy about it. It is simply being open minded and methodical. I do not give a shit about what symbolism models use. I care that it works. I have learned a lot of tricks that make it work for me, and I suck at explaining them. But no, I have never taken psychedelics in my life. The passive aggressive accusation is abusive, toxic and mean behavior. As part of the whole crash that disabled me with a broken neck and back, I had a full psychological exam and am well above average in every respect. That was nearly $10k for the day and expert witness testimony in my personal injury case against the driver that hit me.