The results are in: 2025 will be an Illustrious year.

My poll on what (order) to retrain LoRAs for Illustrious has concluded, and based on it I’ve created a schedule for conversion and already have enough buzz to do 33, giving me at least a LoRA a day till Feb twenty-something unless something drastic happens (contest, on-site trainer goes away, I can’t access the internet etc.). Taking a clear lead is Confused looking anime girls with interrogation marks on their heads which will be converted and posted on the first day of the year (For me. Since I post LoRAs at the start of a Civitai day, it will actually be just after midnight, and thus Jan 2, in GMT time) of the new year. After that I’ll be retraining the various girls from Marvel comics before moving on to Dagoth Ur and and the Camelot software girls before getting everything else. There is however going to be a couple of days where I’ll post an entirely new LoRA, in both PDXL and ILXL flavors (ILXL will be made after I've verified the PDXL version works), in honor of a major anniversary of the source material. Once these surprises are done in a few weeks, I’ll post my full conversion schedule. While I’ve been away I’ve also got a bunch of ideas for stuff I want to try making a LoRA of (15, one of which may require four separate LoRAs to accomplish), which I’ll crosstrain to ILXL the next day, so I’m stocked till mid March.

I’m going to be posting the ILXL conversions as new resources. This is primarily due to site limitations: Filtered search will only find results where the newest version (actually left-most with manually sorted versions) matches the filtered base model (this is a known issue) and the cover card (do these have an actual name?) only lists the newest version’s base model in top left. Since I have the chance I’m also going to altering some tags slightly and occasionally throwing in new training data to address weaknesses in the original version where possible and I want to update the description to reflect that, but per version descriptions aren’t possible. It also gives a me a better view at a glance of what my most popular models are.

My Moon LoRA and Djinn LoRA will not be converted. The first because my self imposed data set limitation of public domain stuff only made it suboptimal at best (it turns out photos of our moon lack the things SD learns off of since they’re a giant mass of grey and more grey) and because acaciahoney did it better by not sticking to my limitation (if someone really, really wants my specific version converted ask and I’ll give you the data set to throw in the trainer yourself since it’s all public domain. I even had it up for DL initially, but it seems that expires after 30 days even when publicly released). The second because it’s a failure for non-Mercury Djinn, I found a bunch of new training data, and wanted to try a version with just the GS1&2 designs for both PDXL and ILXL, but this goes beyond mere conversion.