v0.0.0 (Initial Release)Concept &amp; Description: This LoRA is designed to enforce and control character positioning, spatial alignment, and camera angles/perspectives.Dataset Breakdown (Total: 70 images):<ul><li>Bust Chart / Lineup Profile: 40 images featuring characters aligned in a horizontal row viewed from a side perspective (<code>bust chart</code>, <code>lineup profile</code>, <code>from side</code>).</li><li>Gaijin 4koma (Meme): 10 images based on the iconic reaction meme format.</li><li>Ghost in the Shell Lineup: 10 images capturing the classic lineup aesthetic (<code>ghost in the shell lineup</code>).</li><li>Frontal Lineup: 10 images featuring characters aligned in a horizontal row viewed from the front (<code>lineup</code>).</li></ul>

Concept > Bust chart / gaijin 4koma (meme)

Tweaks. Changes from v0.0.0:<ul><li>Training dataset adjustments (70 → 75 images)<ul><li>Removed 20 images featuring "ghost in the shell lineup" / "lineup", depicting characters lined up side by side from the front.</li><li>Added 20 new images featuring "kirara_jump" / "wanima_photo_(meme)".</li><li>Added 5 additional images from "motteke! serafuku" (Lucky Star).</li></ul></li><li>Training settings<ul><li>Increased dim from 4 to 8.</li><li>Changed the training schedule from cosine 600 to constant 700 + cosine 200.</li></ul></li></ul>dim8/alpha8, AdamW8bit, UNet LR 1e-04, constant 700 + cosine 200(200/300), 512² resolution, 75 images.

Files with SHA256: 01e2cfdd974b37c42c3605838a4ff024a77d25b62c97fa0347d63a067715a071