I usually generate with ComfyUI where I can set the configuration setting down to the third decimal point like 8.998. If I generate down to this resolution while setting the seed to fixed, and all other parameters kept constant, each 0.001 step in configuration will alter the image in very subtle increments. Based on the models I use, this image looks like the eyes or the lower left fire are what the model was adjusting actively at this fine resolution level. Perhaps you already know about adjusting resolution at this scale, and this is indeed the best compromise for results. I figured it was worth mentioning, if you stepped in either direction it might have removed the dual eyes thing.
I usually generate with ComfyUI where I can set the configuration setting down to the third decimal point like 8.998. If I generate down to this resolution while setting the seed to fixed, and all other parameters kept constant, each 0.001 step in configuration will alter the image in very subtle increments. Based on the models I use, this image looks like the eyes or the lower left fire are what the model was adjusting actively at this fine resolution level. Perhaps you already know about adjusting resolution at this scale, and this is indeed the best compromise for results. I figured it was worth mentioning, if you stepped in either direction it might have removed the dual eyes thing.
It’s an interesting style.