左乐/zuo le（明日方舟/arknights）

作者：DSLM

左乐 Zuo Le（明日方舟）· MOSS-Image 角色 LoRA 明日方舟「左乐」的角色 LoRA。蓝发、蓝眼、尖耳、蛇尾，覆盖本体（精零 原皮）、本体（精二 原皮）、皮肤 少年游、皮肤 无息破局、皮肤 音律联觉2025。在 MOSS-Image Base 上训练，建议使用 SFT 模型跑图。 已知问题：疑似因为每套皮肤总步数太少导致衣服细节复刻度差。 已知问题：素材图的发色差距太大，导致生成图的发色不稳定。 【角色触发词】 MoChaZuoLe把触发词放在正向提示词第一行，逐字一致，前后不要加空格或标点，后面换行写画面。 只写触发词就能出角色特征（蓝发、蓝眼、尖耳、蛇尾），但不指定服装；没写服饰时服装会随机，这是角色 LoRA 的正常表现。 【本体（精零 原皮）】 MoChaZuoLe, 1boy, blue hair, blue eyes, pointy ears, snake tail, black cloak, black boots, chinese clothes, sword, vambraces, two-sided cloak, two-sided fabric, belt【本体（精二 原皮）】 MoChaZuoLe, 1boy, blue hair, blue eyes, pointy ears, snake tail, black cloak, black boots, chinese clothes, sword, vambraces, two-sided cloak, two-sided fabric, belt【皮肤 少年游】 MoChaZuoLe, 1boy, blue hair, blue eyes, pointy ears, snake tail, white coat, black trim, green shirt, black belt, grey trousers, fingerless gloves, black boots, folding fan, staff, red ribbon, sword, scroll【皮肤 无息破局】 MoChaZuoLe, 1boy, blue hair, blue eyes, pointy ears, snake tail, black armor, black coat, black pants, black cape, pauldrons, vambraces, black bracers, belt, black boots, sword, scabbard【皮肤 音律联觉2025】 MoChaZuoLe, 1boy, blue hair, blue eyes, pointy ears, snake tail, white coat, blue lining, white shirt, black collar, tassels, blue sash, black pants, ring, black shoes, frog closures【换装】 保留触发词，衣服写细一点；不想要官方服饰就把它放进负面： 负面：black cloak, chinese clothes, vambraces, sword, belt 【推荐参数】

• 底模：MOSS-Image SFT（int8_convrot 或 bf16），Base 也可以

• LoRA 权重：0.9–1.1，从 1.0 起

• 步数：40–60，推荐 50；CFG 3.0（3.0–3.5）

• 采样器：euler + sgm_uniform

【训练参数】

• 底模：MOSS-Image Base（SFT 上测试效果更好）

• 数据：42 张（重复次数已计入）＝官图 5×30＝150，本体同人（服饰一致） 19×10＝190，皮肤同人（服饰一致） 18×20＝360，加权采样 A=700

• 8 epochs / 5600 步（＝A 700 × 8），batch 1，768 bucket

• LoRA rank 16 / alpha 16，LR 1e-4，AdamW + cosine（warmup 100，min 0.1）

• bf16 训练，fp16 导出（约 71 MiB）

• 单卡 RTX 4070 SUPER 12 GB，约 6 小时 42 分，最终 loss 0.6796

• caption 不含外貌词，身份由触发词吸收

Zuo Le 左乐 (Arknights) · MOSS-Image Character LoRA A character LoRA for Zuo Le from Arknights. Blue hair, blue eyes, pointy ears, snake tail; covers Base outfit (Elite 0), Base outfit (Elite 2), Youth journey skin, Silent breakthrough skin, Ambience Synesthesia 2025 skin. Trained on MOSS-Image Base; running it on the SFT model is recommended. Known issue: Clothing details reproduce poorly, likely because each outfit receives too few total steps. Known issue: Hair colour varies widely across the training images, so the generated hair colour is unstable. 【Character trigger】 MoChaZuoLePut it on the first line of the positive prompt, exactly as written — no extra spaces or punctuation — then a newline and your scene. The trigger alone already produces the character's features (blue hair, blue eyes, pointy ears, snake tail), but it does not specify the outfit — with no outfit tags the clothing is random, which is normal for a character LoRA. 【Base outfit (Elite 0)】 MoChaZuoLe, 1boy, blue hair, blue eyes, pointy ears, snake tail, black cloak, black boots, chinese clothes, sword, vambraces, two-sided cloak, two-sided fabric, belt【Base outfit (Elite 2)】 MoChaZuoLe, 1boy, blue hair, blue eyes, pointy ears, snake tail, black cloak, black boots, chinese clothes, sword, vambraces, two-sided cloak, two-sided fabric, belt【Youth journey skin】 MoChaZuoLe, 1boy, blue hair, blue eyes, pointy ears, snake tail, white coat, black trim, green shirt, black belt, grey trousers, fingerless gloves, black boots, folding fan, staff, red ribbon, sword, scroll【Silent breakthrough skin】 MoChaZuoLe, 1boy, blue hair, blue eyes, pointy ears, snake tail, black armor, black coat, black pants, black cape, pauldrons, vambraces, black bracers, belt, black boots, sword, scabbard【Ambience Synesthesia 2025 skin】 MoChaZuoLe, 1boy, blue hair, blue eyes, pointy ears, snake tail, white coat, blue lining, white shirt, black collar, tassels, blue sash, black pants, ring, black shoes, frog closures【Outfit swap】 Keep the trigger and describe the clothes in more detail; if you don't want the official outfit, put it in the negative: Negative: black cloak, chinese clothes, vambraces, sword, belt 【Recommended settings】

• Base model: MOSS-Image SFT (int8_convrot or bf16); Base also works

• LoRA weight: 0.9–1.1, starting from 1.0

• Steps: 40–60, 50 recommended; CFG 3.0 (3.0–3.5)

• Sampler: euler + sgm_uniform

【Training parameters】

• Base model: MOSS-Image Base (tested on Base and SFT)

• Data: 42 images (repeats included) = official art 5×30 = 150 + base-outfit fan art 19×10 = 190 + skin fan art (outfit matches) 18×20 = 360; weighted sampling A = 700

• 8 epochs / 5600 steps (= A 700 × 8), batch 1, 768 bucket

• LoRA rank 16 / alpha 16, LR 1e-4, AdamW + cosine (warmup 100, min 0.1)

• bf16 training, fp16 export (~71 MiB)

• Single RTX 4070 SUPER 12 GB, ~6 h 42 min, final loss 0.6796

• captions contain no appearance words; identity is absorbed by the trigger