<h2 id="update:-happy-new-2026-folks!">Update: Happy New 2026, folks!</h2>I wish you all, a healthy and amazing year! For those who have followed me, a special thank you - for the support, for the occasional kind words, and for the patience. I know I am very sporadic when it comes to uploads, but understand that this AI business is just a hobby, and like one it doesn't take priority. With that being said,I've decided that the first upload for the year would be a checkpoint.Also, several quick As to some FAQs I've been getting lately: Q1: Why did you die? A1: Because I fought for the Greater Good. Q2: Why are you not uploading daily? A2: Because I love wasting time, and time wasting takes a lot of time. Q3: Why did you make a discord server and are not present there (daily)? A3: I banned myself from my own server, because I argued with myself and lost, inadvertently hurting my feelings.Q4: Why is your discord server barren? A4: Honestly, I have no idea. Guess she's too old. Q5: Who is Krivia and will you upload her LoRA(s)? Q5: She's many things - my profile's mascot, an enigma, and, for those who are interested in psychology, my Anima. Jung wrote some interesting theories, go read him, seriously. Also, no, I will not upload any of her LoRAs for various reasons. You'll just have to enjoy her sporadic appearances here or there among some example images of other models. Q6: X LoRA is better than yours. X creator is better than you. A6: Best question. The answer is Yes.<hr />I've long searched for a checkpoint based on IllustriousXL—a fine-tuned SDXL model that I could fully stand behind and use as my go-to. I’ve found a few that came close and have stuck with them for a while, but each one, as impressive as it was, came with a few glaring and frankly frustrating limitations.After many failed attempts to create a cohesive SDXL-based checkpoint of my own, I decided to go all in. What followed was a long cycle of trial and error: dataset curation, Dreambooth training, merging experiments, and layering LoRAs and LoCONs until I arrived at this: v04, the current iteration of what I hope is a versatile, balanced, and usable base model.Why v04? The journey began with me curating a mixed dataset of illustrations and photographs—personally handpicked based on what I look for: artists, poses, shapes, themes, and visual tone. I trained the first Dreambooth version on NoobAI, which I found to be a great base—it blends with different styles more fluidly, converges just as well as Illustrious, and avoids some of the strange quirks that tend to show up in Illustrious models. You probably know the ones I mean—hard to describe, but unmistakable once seen.Here's the evolution:<ul><li>v01 – the initial trained checkpoint on NoobAI using the custom dataset.</li><li>v02 – a merge of v01 with another model trained on an extended version of the dataset (with different repeat weights).</li><li>v03 – a combination of v02, yet another variant trained with tweaked parameters, plus three IllustriousXL-based finetunes.</li><li>v04 – a final merge between v03 and an experimental checkpoint made by merging NoobAI with several of my own LoRAs.</li></ul>I’ve added a trigger word: <code>"pkmerc"</code> — it’s optional. It can improve result stability, but it’s not necessary. Many generations won’t differ much whether you use it or not.One final note: the original base checkpoint that came out of the Dreambooth training (before v01) has the strongest personality and visual signature. It's bold and interesting, but unfortunately doesn’t handle LoRAs very well. I’ve decided to hold it back for now, but I may release it later as a standalone artistic model. Same with versions 01 to 03. If there is any interest or curiosity, I might upload them too.

MercuryXL

Updated some things, kept others, should have an overall same feeling as v04, but still have its own flavor. There is both IllustriousXL and NoobAI in this checkpoint, so LoRAs trained on both will work.I've said it many times, but to reiterate, my goal with this checkpoint or any other I might train is to try and make something more different than the usual "famous" and "known" Illustrious-based checkpoints. What those produce as output, to me, is ugly and bad - smooth, plasticky texture, no linework, simplistic shadows on top of overblown highlights - you've seen one, you've seen them all. Also, the faces and bodies. No. Just no. Too childlike for comfort. Illustrious is biased towards this, yes, so one can't really do much, but most custom merges perpetuate this, making it worse instead of trying to fix it. My unofficial "rule" for testing - if it produces (be it LoRA or especially a checkpoint) good, realistic looking (or as close to it as possible) pubic hair - you know, hair that looks like hair, not a slap of texture-less, volume-less colour - then it's slightly better than the overwhelming majority of custom IL-based merges. Now go and test that with your favourite merge.

Files with SHA256: 007cfa8e3d8e8826c9618e97cfb3915222c70c0627a485c0825a29bd433b0bf1