<h2 id="gypsypunk-v1.0-lk5e1dnqn">GypsyPunk v1.0</h2>GypsyPunk - this will probably undergo a bit more work. It is turning out decent pictures, but still doesn't quite fit what I envision.<h3 id="vikingpunk-v2.1-0tlgmdrfy">VikingPunk v2.1</h3>This is a ffinetune of <a target="_blank" rel="ugc" href="https://civarchive.com/models/660912?modelVersionId=883987">VikingPunk v2.0</a>. There will probably be a new version of VikingPunk with a new dataset before too long since it is my most popular punk style.<h3 id="frostpunk-v1.0-h7cq3ul6z">FrostPunk v1.0</h3>FrostPunk was easy to gather pictures for, but it went through several training sessions of fine tuning the image selection, until it gave the results I wanted. BioPunk v1.0NOTE: uploaded new bioPunk 2025-02-21 21:45 MST<ul><li>Update was trained 'on top' of the existing 1.0 Lora; it added descriptions using the original images. It was trained with 2k steps and 7 Epoch's. If you're counting, that's a total of 4k + 2k = 6k steps and 6 + 7 = 13 Epoch's.</li></ul>bioPunk is another high quality Lora (meaning it can be used with a 1+ strength without loosing any quality - haven't tested the 'top end', I use it at 1). Trigger word is, of course, biopunk. I spent a lot of time gathering / creating the images for this. This version was trained without image prompts, just one overall 'biopunk' trigger word. v2.0WesternPunk is probably my 'highest quality' Lora. I have not tested the top end, but it can easily be run at a full 1 strength w/o any issues.NOTE: I replaced the VikingPunk Lora 9/29/24. It is working pretty well, finally, at .8+. This will be the last 'informal' update to the VikingPunk Lora. Next update will be a version release. Mechanical Arms is now live. Although it is listed as v1, it is V2 quality, and produces great images. Viking Punk is now live. Ii ran this multiple times until I was happy with the results. I also found a big quality issue with augments , and I’m doing another update to augments.Augments is the first Lora in v2.0. This marks quite a deviation, and in my opinion, great improvement in quality. I trained this model myself in ComfyUI using 512, 768, and 1024 datasets, which seems to significantly improve the quality. <h3 id="v1.0-54gwbdjjf">v1.0</h3>The westernpunk, ninjapunk, and vikingpunk are idea's I came upon early in my practice of this hobby. I have kept circling around to them. The westernpunk Lora, in particular, will need future version to work 'fully', but with some other wording, the Lora helps get some pretty great images. I am typically used the full JSON style from my <a target="_blank" rel="ugc" href="https://civarchive.com/models/640908?modelVersionId=719987">punk styles</a> to trigger individual styles.The combination seems to do a pretty good job. The dataset for these Lora's are based upon my own images since I started this hobby about a year ago. I am using <a target="_blank" rel="ugc" href="https://civarchive.com/models/627377?modelVersionId=716879">this workflow</a> to generate the images, and some of the images have used <a target="_blank" rel="ugc" href="https://civarchive.com/models/640908?modelVersionId=719987">these punk styles</a>.

Moxie Cybernetic & Punk Lora's

Done on ComfyUI - new dataset. Ran this and adjusted the input images multiples time. I'm please with the quality improvements. FYI - viking punk will not trigger, must have the underscore viking_punk.

Files with SHA256: 9d583089b520a134f9435ef63373584af3844a5cb3bed81eeb312bd8ec8d986c