Welcome to Forgia, an experimental blending of Illustrious and Pony SDXL models. Leveraging the Merge Block Weighted process within ForgeUI (and harnessing the computational expertise of Gemini to help me work with the numbers), Forgia is an Illustrious-based model fairly deeply merged with Pony architecture models for approximately fifteen generations across a spread of settings, all within my own custom model ecosystem. What began as the FurStein series a few weeks ago has turned into something different enough that, once again, I'm branching it out into something new. As this has been heavily blended with models that do not share its initial dataset, this is a highly experimental model (at its base), that, thanks to included Pony-model data, gives it new textures, lighting, details, backgrounds, costume designs, and other elements, while deviating only moderately from my other recent models. What does Forgia do different than my previous models? A little bit of everything. Try running some of your old prompts through it and see what happens. The new data injections bring out a bit more 2.5D/semirealistic qualities, and it can more faithfully handle some classic prompts that work with Pony models, but it remains more faithful to the Illustrious architecture at the end of the day. What does it struggle with? Without ADetailer, eye details may vary or be poor; this may be from the Pony data, or it may be something else, and will hopefully be improved with a future update. There are occasional floating white spots in some images (I'm pretty confident this is a Pony-related problem, and I haven't found a way to completely remove it from the model just yet) that can be ameliorated somewhat in the negative prompt. What are my plans for it? As with all of my models, this is entirely an experimental journey, in this particular case, a journey to push the limits of just how deeply Illustrious and Pony models can be merged, by reaching as close as possible to the muddy middle ground where the images generated simply break. Hopefully, I can ease it into being a bit more stable, but you should expect a variety of results, potentially some unexpected ones, with this Forgia. That said, if you enjoy seeing creative results that don't feel like they're all practically identical, you might enjoy this one as well. If you come across anything that seems broken, feel free to leave a note about it in the comments, and if possible, I will keep an eye out for it as my experiments move forward, and I'll try to target a fix for it (though, it may not be possible with the process I'm using, and the relatively limited experience I have from brute-force experimentation.)

Forgia

Initial version. As always, no pesky begging for your finances, no early access nonsense. Comes with sdxl_vae.safetensors baked in, though you may use a VAE of your preference with it as well. Illustrious LORAs are not guaranteed to work, as I don't have the resources to test every one of them with it, and blending Pony data into the mix may result in unpredictable outcomes.

Files with SHA256: 7dcbbb9a466de8484b6f8f523f7f7cf5eea1e765029808e466fc46c0579374be