<h1 id="sandrone-2.5d-realism">Sandrone — 2.5D Realism</h1><p>So, this is my <strong>Sandrone LoRA from Genshin Impact</strong>, trained on <strong>HassakuXLIllustrious</strong>.</p><p>The main focus of this LoRA was to push Sandrone towards a <strong>2.5D realistic/semi-realistic appearance</strong> instead of keeping her purely in the usual anime style.</p><p>Because of that, the identity does shift a bit compared to her original Genshin appearance. I intentionally went for a more realistic interpretation of her rather than trying to make a perfect 1:1 recreation of the original model. The facial structure, proportions and overall appearance can therefore look somewhat different depending on the prompt and LoRA weight.</p><p>The idea was basically:</p><p><strong>Sandrone → anime character → 2.5D interpretation</strong></p><p>while still keeping her recognizable as Sandrone.</p><h3 id="trigger-word">Trigger Word</h3><p><code>Sandrone</code></p><h3 id="recommended-weight">Recommended Weight</h3><p>Around <strong>0.7–0.9</strong> is a good starting point.</p><p>Lower weights can give you a more natural 2.5D interpretation, while higher weights will push the character features harder.</p><h3 id="base-model">Base Model</h3><p><strong>HassakuXLIllustrious</strong></p><h3 id="a-few-things-about-the-lora">A few things about the LoRA</h3><p>This was trained specifically with the <strong>2.5D realism goal</strong>, so don't expect the result to look exactly like Sandrone's original Genshin artwork.</p><p>It works particularly well when you give the prompt enough information about the character's appearance instead of relying entirely on the trigger word.</p><p>I also recommend experimenting with different weights depending on the checkpoint you're using, since the exact result can vary quite a bit between models.</p><p>Have fun with it.</p>

Files with SHA256: aa8614e8d2d23d731da545b238809a24fd484c5afc9267b2f4fba261c9173430