This LoRA must not be used to generate NSFW content.Expand this post for the recommended Anima parameters.Suggested tags for consistent results:<pre><code>military uniform, green military uniform, long sleeves, high collar, epaulettes, belt, green pants, black boots
</code></pre>Recommended settings:<ul><li>DPM++ 2M </li><li>Karras is prefered </li><li>Sampling Steps: 25-35 </li><li>CFG Scale: 5-6 </li><li>Model Weight: 0.8-1 (tested with 0.8) </li><li>Clip Skip: 2 </li></ul>This LoRA was trained on a small, mixed-quality dataset.As a personal preference, I generally use ADetailer for the faces of any generation, this model gave me some issues with the face. I would recommend using a face ADetailer.I have had success with multiple Checkpoints. Realistic checkpoints tend to age up and change body structure.<hr />Anima Version:<pre><code>tanya_kurba, blonde hair, ahoge, poytail, blue eyes, </code></pre><pre><code>green military uniform, epaulettes, belt, green pants, black boots,</code></pre>If you want to add her military cap, replace ahoge and poytail with peaked cap.I've had mixed results while using quality tags. Here are a few that I've tested and found to work across multiple checkpoints.Generally okay to use with any checkpoint:<pre><code>masterpiece, best quality, score_9, score_8, score_7, absurdres, highres, newest, </code></pre>Best with higher-detail checkpoints:<pre><code>hyper realism, photorealistic,</code></pre>The realism tags improved overall quality while preserving the anime style.The lighting was the most notable difference when using these tags.Recommended settings:Sampling Steps: 15-35CFG Scale: 3.5-4I would love to see what you create with this LoRA. If you have any feedback, I'd be be interested in hearing it.

Tanya Degurechaff - [youjo senki: saga of tanya the evil]

This version is trained on the same dataset as the Illustrious version, but using Anima.

Files with SHA256: afda68eb0aef6aad97c777d5894d0e68ea8b6ce4dff2d680f483b06174221866