ModelsLab/Astigmatism (Formerly 'Semantic Shift') - Astigmatism 0.2

This one completely changes the data set and moves from the initial experiment to a more directed effect. So, for example, whereas "a mona lisa Kudrow" would just produce images of the Mona Lisa (which is overtrained in base SDXL), with this LoRa, you actually get Lisa Kudrow/Mona Lisa hybrid. It's worth noting that this is emergent behavior ie I did NOT train against a mona lisa prompt.In fact, I trained against 5 prompts, focusing on the MOST overtrained images I know of, and then trying to associate them with images that actually, semantically, and diversly, represent the meaning of that overtrained phrase. The side effect is that the model appears to, in many cases, be broadly more intelligent.