Furry/anthro artist <a target="_blank" rel="ugc" href="https://e621.net/artists/18891">Dredjir.</a> Made for Anima.<hr /><h3 id="recommended-model-and-parameters">Recommended Model and Parameters</h3>Works best with <a target="_blank" rel="ugc" href="https://civarchive.com/models/2356447/rdbt-or-anima">RDBT Anima Checkpoint.</a>Sampler and Scheduler: Euler A Simple.<hr /><h3 id="recommended-weights">Recommended Weights</h3>V2.5: Doesn't need much messing around with weights, but do change it if your images come out distorted.V2.1: Weigh between 1-1.3 for flexibility and accuracy.V2 and V1.8: Weigh between 0.8-1 for best results.V1.5 and V1: Weigh between 0.6-1 depending on how deformed or deep fried your images are.<hr /><h3 id="helpful-tips">Helpful Tips</h3><ul><li>When generating feet in general, it is usually partially obscured, either hidden by an object or one feet is hidden by the other feet, so adding <code>partially obscured feet</code> in negative prompt typically helps.<ul><li>Note: V2.1 and later versions don't have as much problems with this, but you can still try it if it still persists.</li></ul></li><li>If watermarks and dialogue appear, even though you didn't want it, you can try adding <code>watermark, speech bubble</code> in negative prompt (I already removed all artist watermarks from my dataset, however, your checkpoint and other Loras may be forcing unrelated watermarks onto the image anyways).</li></ul><hr /><h3 id="version-info">Version Info</h3><code>V2.5</code> is retrained again, but with different settings. It is much larger in file size, but way more accurate to original style.<code>V2.1</code> is retrained locally and is smaller in file size compared to previous versions. It has a better balance between prompt and style adherence.<code>V2</code> resembles a lot of the artist's own style and it follows prompts decently but it can still struggle with words or concepts.<code>V1.8</code> is one of the versions similar to the shading that you can get.<code>V1.5 More Shaded</code> is still the same as V1.5 but it has more of a 2.5D look. It does tend to look more like generic furry AI slop.<code>V1.5</code> contains smaller dataset and still rough.<code>V1</code> represents the overall rough artstyle, but it does seem to struggle with artist watermarks and signatures.<hr />If needed, DM me and I will take this model down upon request.Have fun!

Dredjir Style (Anima)

Retrained dataset with different optimizer and increased network rank.Significantly more accurate but larger in file size.

Files with SHA256: 6ce815d882d39da18ad0379f7131a88f4bd2552345811082dd84eabd19ad0e38