I am making this announcement to make my official entry into XL model training official, which is why I am here:
- I want to upgrade all my possible models to XL this year: I know that many people will still prefer 1.5, I was also reluctant to change at first, but I think that with pony and turbo/lightining there are enough incentives to make the change even if there is uncertainty about the release of SD3, even if it comes out we will have to wait for a big finetune of the level of Pony or Animagine, which will surely be at the end of the year or early next year, there is time and millions of possible models still to be made in XL.
- My prompting has always left much to be desired, my plan was always to upload the maximum amount of models in the shortest time possible (because I created and generated with google colab and its limits) without really worrying about the quality, I want to change this now that I have my own team I want to be able to dedicate more time to it, and not leave you again with only two or three example images in each model, my top priority at this moment will be to stop depending on google also for the XL model training.
- One of my qualities has always been my perseverance and search for datasets in all kinds of nooks and crannies of the internet, I want to keep coming up with models that are little known or of little use to others, I believe that in the niche is the key to the loras, let's make civitai like the library of Alexandria, and possess all the possible knowledge in the form of re-training.
- One of the reasons I got burnt out and stopped uploading models despite having them trained was because of the unhealthy approach I had, an approach that constantly gave me bottlenecks, I think it's time to adopt a new approach more open, collaborative and slower, now that civitai has changed and allows things like rewards, I'm going to delegate more to other people, (thanks to all of you who gave me your buzz, I'm thinking of spending it wisely).
- That said, given the magnitude of what I have set out to do:
- I give permission to anyone who wants to train a model I made in SD 1.5 in XL (not that you had to ask me before xd).
- I will give by batches the training datasets of my models in 1.5, to the people who contact me, for the moment I will release 60 approx. in a list that I will leave in another post.