SVD Distilled Version (Rank 128) for FLUX.2This is an SVD-distilled version of our massive full-weight experimental merge that combined Van Gogh (Rank 2048) and Aivazovsky (Rank 1536). By applying Yang's SVD matrix decomposition formula, we compressed the original 23.25 GB file down to a compact 1.5 GB while preserving over 90% of the core style energy.Model Characteristics:<ul><li>SVD Compression: Cleans up high-rank redundant noise while retaining Aivazovsky's dramatic translucent water lighting and Van Gogh's signature sky dynamics and impasto textures.</li><li>Low-Weight Flexibility: At low model strength (<code>0.01</code> – <code>0.05</code>), it injects master-level color, mist, and atmospheric lighting into photorealistic/cinematic prompts without forcing a heavy oil-painting texture.</li></ul><blockquote>Links &amp; Resources:<ul><li>Full weights, alternative checkpoints, and native workflows are available on my Hugging Face.</li><li>Dataset filtering scripts and training tools are published on my GitHub.<a target="_blank" rel="ugc" href="https://github.com/OrakulStudio">https://github.com/OrakulStudio</a></li><li><a target="_blank" rel="ugc" href="https://huggingface.co/OrakulStorm">https://huggingface.co/OrakulStorm</a></li></ul></blockquote><h3 id="recommended-parameters">Recommended Parameters</h3><ul><li>Base Model: FLUX.2</li><li>Model Strength (LoRA Weight): <code>0.05</code> (use <code>0.5</code> for full painterly style)</li><li>CLIP Strength: <code>1.0</code></li><li>Sampler: <code>UniPC</code></li><li>Scheduler: <code>SGM Uniform</code></li><li>Steps: <code>35</code></li><li>Triggers: r2048f2dvangog, r1536f2daivaz</li></ul>

Files with SHA256: ef2d7ae11ca038e5634bdbc25050053984c24b173716206de42c5cb804e636ed