Nanami Jinnai - LoRA Training Report (Illustrious-XL)💻 Hardware &amp; Environment<ul><li>GPU: NVIDIA GeForce RTX 5060 Ti (16GB VRAM)</li><li>Training Suite: Kohya_ss GUI (Local deployment)</li></ul>💿 Base Model &amp; Dataset<ul><li>Base Model: Illustrious-XL v0.1 (Checkpoint)</li><li>Dataset Size: 18 high-quality retro anime screenshots (Manually isolated &amp; cropped)</li><li>Aesthetic Style: 1990s vintage anime cel-shading (Pioneer LDC style)</li></ul>⚙️ Training Parameters<ul><li>Dataset Repeats: 33</li><li>Batch Size: 4</li><li>Max Train Epochs: 3</li><li>Total Optimization Steps: 450 (150 steps per epoch)</li><li>Resolution: 1024x1024 (Native Square Buckets)</li><li>Clip Skip: 2</li></ul>📦 Network Config &amp; Size<ul><li>Network Rank (Dimension): 32</li><li>Network Alpha: 16</li><li>File Format: <code>.safetensors</code></li></ul>🧠 Solvers &amp; Optimizations<ul><li>Optimizer: AdamW8bit (Stable &amp; VRAM efficient)</li><li>Gradient Checkpointing: Enabled (VRAM usage locked at a stable ~10GB)</li><li>Mixed Precision: fp16</li></ul>🎯 Recommended Generation Settings (ComfyUI / WebUI)<ul><li>Best File Version: Epoch 1 (<code>-000001.safetensors</code>) — Offers the cleanest anatomy and absolute anti-cloning stability.</li><li>LoRA Weight: 0.85 – 1.0</li><li>CFG Scale: 5.0 – 5.5</li><li>Sampler: <code>euler_ancestral</code> (Euler A) with <code>normal</code> scheduler</li><li>Recommended Resolution: 896x1152 (Portrait) or 1024x1024 (Square)</li></ul>🟩 Core Activation Tags (Danbooru Style)<ul><li><code>GTML_Nanami_Jinnai, 1girl, solo, anime screenshot, short hair, brown hair, repicado hair, hairband, black hairband, brown eyes, 1990s anime style, retro anime</code></li><li>School Uniform Outfit: <code>school uniform, yellow cardigan, unbuttoned cardigan, yellow blazer, collared shirt, white shirt, necktie, red tie, green skirt, short skirt, pleated skirt</code></li></ul><hr />

Files with SHA256: 4e2e8827906a847f2db597939a9b3fe74dd3b92b327f79f0fcdef308845de7a3