<h2 id="doro-epic-airbrush-soft-gradients">DORO EPIC AIRBRUSH - Soft Gradients</h2><hr /><h3 id="versions">Versions</h3><ul><li>v1 | <code>DEA_SOFT_GRAD_1</code> - Initial release</li><li>v2 | <code>DEA_SOFT_GRAD_2</code> - more concentrated and powerful than v1</li></ul><hr /><h3 id="compatibility">Compatibility</h3><ul><li>Illustrious XL 🟢 v1 full | 🟢 v2 full</li><li>Pony XL 🟡 v1 partially | 🟢 v2 full</li></ul><hr /><h3 id="quick-start">Quick Start</h3><hr />🏷️Trigger v1: <code>DEA_SOFT_GRAD_1</code>🏷️Trigger v2: <code>DEA_SOFT_GRAD_2</code><hr />⚠️ High-offset LoRA - effective range starts at 2.0+, not the usual 0.5–1.0<hr />🏆 Sweet spot v1:<ul><li><code>0.7–1.9</code> - subtle, barely visible</li><li><code>2.0–3.0</code> - effect kicks in, full style ⭐</li><li><code>3.0+</code> - overpowered, style compression</li></ul><hr />🏆 Sweet spot v2:<ul><li><code>0.7-1.0</code> - subtle atmosphere, clean polishing range</li><li><code>1.5-2.5</code> - full effect: dramatic lighting, deep shadows, vortex glow ⭐</li><li><code>3.0+</code> - overpowered, latent saturation, loss of fine detail</li></ul><hr /><h3 id="description">Description</h3>📸 Dataset: 15 abstract airbrush gradient crops - no objects, pure tonal transitions and color blending. Trained at 768px, 600 steps, AdamW8bit, cosine_with_restarts.✨ Emergent effects:<ul><li>Gradient surfaces - clouds, smoke, fog, fire, atmosphere: deep analog painterly quality, as if airbrushed on paper</li><li>Smooth surfaces - skin, plastic, metal: surface-blur-like effect, removes micro-noise and texture artifacts, evens gradients</li><li>Object edges - sharpened and stylized, more "painted" feel</li><li>Background-first - at moderate weights affects mainly background; at high weights touches subjects too</li></ul>⚠️ Side effect: Smoothing suppresses fine texture (pores, grain, fabric). Not ideal when texture detail is the goal. Workaround: generate smooth, then add noise + slight blur in Photoshop.💡 Bonus use: Pre-upscale prep - smooths surfaces and reduces artifacts for a cleaner upscale input.<hr /><h3 id="what-happened-under-the-hood">What happened under the hood</h3>This LoRA was trained on abstract gradient crops with no recognizable objects - and that turned out to be the key.The model couldn't learn any specific object, so it learned pure rendering principles: how to blend tones, transitions, and light. When applied, it rewrites the model's rendering language across all volumetric, gradient-by-nature subjects - clouds, smoke, fire - because those subjects are gradients at their core.This is a case of spontaneous feature disentanglement: an abstract dataset forced style and content to separate. The result is a universal style modifier, not a content LoRA - similar in principle to implicit style-content separation described in B-LoRA (ECCV 2024).B-LoRA paper:<a target="_blank" rel="ugc" href="https://arxiv.org/abs/2403.14572">https://arxiv.org/abs/2403.14572</a><hr /><span style="background-color:rgb(21, 21, 23);color:rgb(249, 250, 251);font-family:quote-cjk-patch, Inter, system-ui, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen, Ubuntu, Cantarell, &quot;Open Sans&quot;, &quot;Helvetica Neue&quot;, sans-serif;font-size:16px">❤️ Artificial Inspiration by DORO

DORO EPIC AIRBRUSH - Soft Gradients

Files with SHA256: 47f80db511240b8da50deeb790a1439ed77dbfe738c071cafbce74dd5c04ce9b