Illuminate is a bidirectional SDXL concept LoRA that controls the apparent illumination of an image while attempting to preserve its underlying subject, composition, and scene semantics. Negative weights progressively reduce illumination and visibility; positive weights increase illumination, exposure, and the apparent amount of light present in the scene.Unlike a simple post-processing brightness adjustment, Illuminate operates during generation, allowing the model to reinterpret lighting in a scene-aware manner.<h3 id="range">Range</h3>Recommended experimental range: <code>-4.0 → +4.0</code> IntentThe intent is not merely to make pixels darker or lighter. Illuminate is intended to provide generative control over the semantic concept of how illuminated a scene is.Ideally, that means it can alter things such as apparent light availability, exposure, shadow depth, highlight prominence, visibility of surfaces, and the model's interpretation of the lighting environment, while leaving unrelated concepts comparatively stable.Negative direction (<code>&lt;0</code>) — progressively reduced illumination, deeper shadows, reduced visibility and darker scene interpretation. At stronger negative weights (&lt; -3), information may become obscured or the model may begin interpreting the scene as intrinsically dark rather than merely less illuminated.Neutral (<code>0</code>) — base checkpoint behavior with no Illuminate intervention.Positive direction (<code>&gt;0</code>) — progressively increased illumination and visibility, brighter surfaces and stronger apparent light contribution. At high positive weights (&gt;3), the model may move beyond plausible illumination toward overexposure, flattened contrast, blown highlights, or semantic reinterpretation of the lighting environment.Performance: Illuminate produces a generally progressive bidirectional change in scene illumination, with the most compositionally conservative results expected near the center of its range. Stronger weights increase conceptual intervention and may introduce secondary changes to contrast, color, lighting geometry, environmental interpretation, or scene content. Extreme weights are primarily useful for evaluating the limits and semantic structure of the learned direction.

Files with SHA256: 36db197a2d40bfe423714554be19e9822848ef9dc65315caa61e37f247637b12