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)

Slightly bigger dataset, but it is one of the most closest to Dredjir's artstyle (shading) on this Lora model for now.All watermarks or unnecessary dialogue are cut out and blended seamlessly, resulting in mostly a watermark-free model.

Files with SHA256: d8214908e0e0306cb867ab79bd51bb231ba9c78e60071b829b064604e3b0e915