# Chimera LoRA - Civitai Upload Description## 中文说明这是一个用于生成机械奇美拉少女的 SDXL LoRA。核心概念是：人形女性上半身、机械蜘蛛式下半身、多节机械足、外骨骼装甲、胸部与小腹区域连接的管线，以及偏冷淡、空洞的表情。这个 LoRA 适合生成机械蜘蛛娘、赛博外骨骼少女、武器挂载型机体少女，也适合测试不同武器配置，例如狙击枪、突击步枪、机枪、手枪、刀具或双持武器。灵感来源于游戏《群星》中的机械同化斗士：被傀儡化后失去自我意识的有机体，最终形成了半人半机械的嵌合体存在。### 触发词```textchimera```### 推荐权重默认推荐：```text0.80 - 1.20```### 仅触发词测试```textchimera```### 推荐额外 Prompt基础强化：```textchimera, humanoid upper body, mechanical spider lower body, segmented spider legs, jointed metal legs, armored spider abdomen, chest cables, abdomen cables, expressionless, empty eyes```武器测试：```textchimera, mechanical spider lower body, weapon mounts, holding sniper rifle, segmented spider legs, chest cables, expressionless, empty eyes``````textchimera, mechanical spider lower body, holding assault rifle, jointed metal legs, side armor panels, chest cables, abdomen cables``````textchimera, mechanical spider lower body, gatling gun, machine gun, multiple mechanical legs, weapon mounts, empty eyes``````textchimera, mechanical spider lower body, holding knife, handgun, dual wielding, mechanical exoskeleton, chest cables```如果生成结果偏黑白，可以加入内容向的颜色锚点：```textfull color, blue armor accents, red sensor lights, colored mechanical parts```### 使用建议只写 <code>chimera</code> 应该能召唤基础概念；如果想稳定得到机械蜘蛛下半身、胸腹管线和空洞表情，建议加入上面的基础强化词。不建议在测试时加入过多画风词、线稿词、黑白词或高对比词。这个 LoRA 本身有大量金属装甲结构，过强的线稿、高对比和阴影提示词可能让画面变灰或接近黑白。## English DescriptionThis is an SDXL LoRA for generating a mechanical chimera girl. The core concept is a humanoid female upper body combined with a mechanical spider-like lower body, segmented metal legs, exoskeleton armor, cable connections around the chest and lower abdomen, and a cold expressionless face with empty eyes.It is suitable for cyborg arachne characters, weapon-mounted mechanical girls, and different weapon configurations such as sniper rifles, assault rifles, machine guns, handguns, knives, or dual-wield setups.Inspired by the Machine Assimilator warriors from the game Stellaris: organic beings who have lost their self-awareness after being turned into puppets, forming half-human, half-machine chimeric entities.### Trigger Word```textchimera```### Recommended WeightRecommended range:0.80 - 1.20```### Trigger-Only Test```textchimera```### Recommended Extra PromptsCore reinforcement:```textchimera, humanoid upper body, mechanical spider lower body, segmented spider legs, jointed metal legs, armored spider abdomen, chest cables, abdomen cables, expressionless, empty eyes```Weapon tests:```textchimera, mechanical spider lower body, weapon mounts, holding sniper rifle, segmented spider legs, chest cables, expressionless, empty eyes``````textchimera, mechanical spider lower body, holding assault rifle, jointed metal legs, side armor panels, chest cables, abdomen cables``````textchimera, mechanical spider lower body, gatling gun, machine gun, multiple mechanical legs, weapon mounts, empty eyes``````textchimera, mechanical spider lower body, holding knife, handgun, dual wielding, mechanical exoskeleton, chest cables```If the result becomes too monochrome, add content-based color anchors:```textfull color, blue armor accents, red sensor lights, colored mechanical parts```## Training Parameters- Base model: <code>chenkinNoobXLCKXL_v03.safetensors</code>- Architecture: SDXL LoRA- Network module: <code>networks.lora</code>- Dataset: <code>64</code> images- Dataset folder: <code>3_chimera</code>- Repeats: <code>3</code>- Total train images per epoch: <code>192</code>- Epochs: <code>12</code>- Total steps: <code>2304</code>- Batch size: <code>1</code>- Resolution: <code>1024x1024</code>- Bucketing: enabled- Bucket range: <code>512 - 2048</code>- Bucket no upscale: enabled- Network rank/dim: <code>64</code>- Network alpha: <code>32</code>- Network dropout: <code>0.05</code>- Learning rate: <code>8e-5</code>- U-Net LR: <code>8e-5</code>- Text encoder LR: not used- Scheduler: <code>cosine</code>- Warmup steps: <code>100</code>- Optimizer: <code>AdamW8bit</code>- Mixed precision: <code>fp16</code>- Clip skip: <code>2</code>- Caption shuffle: enabled- Keep tokens: <code>1</code>- Max token length: <code>225</code>- Caption dropout rate: <code>0.10</code>- Caption tag dropout rate: <code>0.05</code>- Min SNR gamma: <code>5</code>- Noise offset: <code>0.03</code>- Cache latents: enabled- Cache latents to disk: enabled- Gradient checkpointing: enabled- Flip augmentation: disabled

Files with SHA256: 77fe80a3242b69711595cd5f9ad3f33e9f611721cf394c55e9a24d34725c66ba