<h3 id="in-memory-of-leonardo">In Memory of Leonardo</h3><p>A style LoRA built entirely from my own urban exploration photography, years of abandoned villas, chapels, workshops and forgotten rooms, condensed into one look.<br /><br />The name is a quiet tribute. Sfumato, the soft transition from light into shadow, is what defines these places for me: a single beam through a broken window, dust in the air, and everything else slowly dissolving into darkness.</p><h3 id="trigger-word"><strong>Trigger word</strong></h3><pre><code>X4LSFUMATO</code></pre><p><br /><strong>Style prefixes</strong> (from the three training sets)</p><pre><code>X4LSFUMATO, dark cinematic color grading, deep shadows, heavy vignette</code></pre><pre><code>X4LSFUMATO, warm muted film tones, soft faded colors</code></pre><pre><code>X4LSFUMATO, natural warm light, documentary photo</code></pre><h3 id="recommended-settings"><strong>Recommended settings</strong></h3><p>- Base model: Krea 2 Turbo (trained on Krea 2 Raw)</p><p>- Strength: 1.0 – 1.4, up to 2.5 for an extreme painterly look</p><h3 id="tips"><strong>Tips</strong></h3><p>- Keep prompts short and factual: the place, the objects, where the light comes from, the camera angle. Long style or lighting descriptions override the LoRA.</p><p>- One light source works best. A single beam onto the main subject with the rest of the room falling into shadow is exactly what it was trained on.</p><p>- For the punchy dark look, add: teal blue shadows and warm orange highlights, high local contrast</p><p>- It also adds decay to scenes that are not abandoned at all, but stays subtle when the prompt clearly describes an active place.</p><h3 id="known-limitations"><strong>Known limitations</strong></h3><p>- Torn fabric (curtains, tarps) sometimes gets unnaturally round holes.</p><p>- On surfaces like signs or license plates the trigger word can show up as text. Avoid those in the prompt or retry with another seed.<br /></p><p>Dataset: 72 of my own photographs, captioned by hand and with a local vision model, rank 32.<br /></p><p>A MOSS version is planned.</p>

In Memory of Leonardo

<p>Codex X – MOSS version.</p><p>Trained with AI Toolkit on the same 72 urbex photographs as the Krea 2 version, 800 steps. Recommended with the MOSS-Image SFT 8-step DMD LoRA, strength 0.8 – 1.2. Adds decay, grit and the teal-orange contrast of the original photos to MOSS's clean look.</p>

Files with SHA256: 9232fca4792a80f23a93ebed902b49d990a6dc951b85b7ece1a78ebf83a8dc92