Photographic Scene Coherence

PSC_large – Texture Refinement (v1.6)PSC_large is an incremental refinement of PSC V1.5, focused on improving surface textures, microdetail, the interaction between materials and light, and character interaction. Rather than a major overhaul, this update builds on the existing model with a much larger training dataset of images.Key improvements<ul><li>More consistent surface textures: Wood grain, weathered wood, stone and glass surfaces appear more differentiated and coherent, with fewer irregular or clumped structures.</li><li>Improved hair definition: Individual strands are more clearly distinguishable, particularly in dark hair, where V1.5 tends to produce clumping.</li><li>More convincing light and material response: Highlights, shadows and reflections often provide better material separation and a stronger sense of volume.</li><li>Reduced unwanted particles: The tendency to generate unintended particle-like details, such as sweat-like specks, is noticeably reduced.</li><li>More nuanced detail rendering: Fine structures are often easier to read without relying as heavily on exaggerated local contrast.</li><li>More expressive character interaction: Compared to V1.5, PSC_large shows a stronger tendency to intensify interactions between characters. This can manifest in more expressive body language, more pronounced facial expressions and a greater sense of emotional or physical engagement between characters. This is not a guaranteed effect in every generation, but it is a tendency accounted for in the dataset and part of the model's overall refinement.</li></ul>Usage notesPSC_large is more sensitive to LoRA strength than V1.5. While V1.5 can often be used reliably at 0.8, PSC_large benefits from testing weights between 0.4 and 0.8, depending on the image and desired effect.In short: PSC_large is not a major update, but rather a step forward toward more coherent textures, better-defined hair, more convincing material rendering and potentially more expressive character interactions. The improvements vary by scene and are best understood as a consistent refinement rather than a dramatic transformation.

Files with SHA256: 722c54c48728a525a76dcb9ba6e23e890795089ecae703c8e46b6c3f61a63abb