kongbai-84/soultide_lora_Anima

Soul Tide LoRA Anima / 灵魂潮汐 LoRA Anima

This repository contains a character LoRA collection for Soul Tide, with additional included original characters and supporting prompt-management files. 本仓库包含《灵魂潮汐》角色 LoRA，并包含协议中列明的原创角色以及配套提示词管理文件。

The LoRA file is 灵魂潮汐-V3.safetensors. LoRA 模型文件为 灵魂潮汐-V3.safetensors。

Training Information / 训练信息

The model was trained with 1024 and 1344 training resolutions. 模型使用 1024 与 1344 训练分辨率。

Both network_dim and network_alpha were set to 48. network_dim 与 network_alpha 均设置为 48。

Layers 19 through 27 were excluded from training. 训练时剔除了第 19 至第 27 层。

The current model fit is still somewhat insufficient. 目前模型拟合度仍有些不足。

Because personal time and energy were limited, the training dataset could only be expanded to a limited extent. 由于个人精力有限，训练集扩充幅度也比较有限。

Style overfitting is still present, so generated images may retain a noticeable learned visual style. 画风过拟合问题依然存在，因此生成结果可能带有较明显的训练画风倾向。

Repository Files / 仓库文件

灵魂潮汐-V3.safetensors is the LoRA model file. 灵魂潮汐-V3.safetensors 是 LoRA 模型文件。

提示词.html is the standalone SD Character Prompt Manager Pro web page. 提示词.html 是可独立使用的 SD 人物提示词管理器 Pro 网页。

提示词.json contains the character tree, prompt tabs, classifications, and Chinese translation dictionary used by the web page. 提示词.json 包含网页使用的人物树、提示词标签页、分类信息和中文翻译词典。

The current JSON includes Anima and Illustrious prompt collections with 4,688 translation entries. 当前 JSON 包含 Anima 与 Illustrious 两套提示词分类，并收录 4,688 条翻译词典数据。

The Anima collection currently contains 103 characters and 397 prompt tabs. Anima 分类目前包含 103 个人物和 397 个提示词子内容。

The Illustrious collection currently contains 114 characters and 400 prompt tabs. Illustrious 分类目前包含 114 个人物和 400 个提示词子内容。

训练集-灵魂潮汐.7z contains the shared training dataset archive. 训练集-灵魂潮汐.7z 是随仓库提供的训练集压缩包。

使用协议.txt contains the required usage terms for the LoRA, included characters, derivatives, and generated images. 使用协议.txt 包含 LoRA、所含角色、衍生训练以及生成图片的使用条款。

Prompt Web Page Quick Start / 提示词网页快速开始

Open 提示词.html with a recent Chrome or Edge browser. 请使用较新版本的 Chrome 或 Edge 打开 提示词.html。

The page uses the browser File System Access API for opening and overwriting local JSON files. 该网页使用浏览器 File System Access API 打开和覆盖保存本地 JSON 文件。

Click 打开文件 and select the downloaded 提示词.json. 点击“打开文件”，然后选择下载好的 提示词.json。

Use the model tabs at the top of the sidebar to switch between Anima and Illustrious data. 使用侧边栏顶部的模型标签页切换 Anima 与 Illustrious 数据。

Use the search box to find characters or folders, then select a character from the tree to view its prompt tabs. 使用搜索框查找人物或文件夹，再从树形列表中选择人物并查看其提示词子内容。

The layout button switches between vertical and horizontal layouts, and the divider can be dragged to resize the navigation area. 布局按钮可切换竖版与横版界面，中间分隔条可拖动调整导航区域大小。

Editing Prompts / 编辑提示词

Select a character and choose a prompt tab such as a costume, appearance, or other child entry. 选择人物后，再选择服装、外观或其他类型的提示词子内容标签页。

Edit prompt text directly in the text area, using commas to separate tags and line breaks to organize groups. 可直接在文本框中编辑提示词，使用逗号分隔标签，并用换行整理不同分组。

The editor automatically converts Chinese commas to English commas and removes repeated commas. 编辑器会自动将中文逗号转换为英文逗号，并清理连续重复的逗号。

Pasted prompt text is automatically given a trailing comma when needed. 粘贴提示词后，网页会在需要时自动补充末尾逗号。

Use the + tab to create a new child entry, double-click a tab to rename it, drag tabs to reorder them, or use the close icon to delete one. 使用“+”标签页可新建子内容，双击标签页可重命名，拖拽可排序，关闭图标可删除。

You can create folders and characters, delete selected tree entries, sort folders before characters, and drag entries to reorganize the tree. 你可以新建文件夹和人物、删除选中的树节点、按文件夹优先排序，并通过拖拽重新整理目录。

Tag Preview and Copying / 标签预览与复制

Each prompt is rendered as individual tags below the editor. 每段提示词都会在编辑框下方渲染为独立标签。

Click a tag to copy its English text, drag it to reorder tags, or use the close icon to remove it. 单击标签可复制英文内容，拖拽可调整顺序，关闭图标可删除标签。

Double-click the English text to edit the original tag, or double-click the Chinese text to add or update its translation. 双击英文可修改原始标签，双击中文可添加或更新对应翻译。

Use the line-break icon on a tag to insert a new line immediately after it. 使用标签上的换行图标可在该标签后插入新行。

When the pointer is hovering over a tag, the default shortcuts are C for copy, D for delete, Q for editing English, W for editing Chinese, and N for inserting a line break. 鼠标悬停在标签上时，默认快捷键为 C 复制、D 删除、Q 编辑英文、W 编辑中文、N 插入换行。

These shortcuts can be changed in Settings and are designed to work while a Chinese input method is active. 这些快捷键可在设置中修改，并针对中文输入法状态进行了兼容处理。

Use 复制全部 to copy the current prompt tab, with an optional setting to append a comma automatically. 使用“复制全部”可复制当前提示词子内容，并可在设置中控制是否自动追加逗号。

Batch Copy and Classification / 批量复制与分类

The batch panel can copy all child entries for the current character or exclude the entry named 道具. 批量复制面板可复制当前人物的全部子内容，也可排除名为“道具”的子内容。

It can also copy only facial features, facial plus upper-body features, or the current prompt with shoe-related tags removed. 它还可以仅复制面部特征、面部加上半身特征，或复制已排除鞋履标签的当前提示词。

Custom selection mode supports choosing multiple characters and selecting specific child entries for each character before copying. 自定义选择模式支持同时选择多个人物，并为每个人物指定需要复制的子内容。

Face, upper-body clothing, and footwear classifications can be edited manually for each character prompt tab. 每个人物的提示词标签页都可以手动编辑面部、上半身服装和鞋履分类。

Optional AI classification can fill missing categories while preserving the original English tags. 可选的 AI 分类功能可以补充缺失分类，并保留原始英文标签。

Optional AI Features / 可选 AI 功能

Open Settings to configure an OpenAI-compatible API key, endpoint, model name, thinking mode, and concurrency. 打开设置可配置兼容 OpenAI 格式的 API Key、接口地址、模型名称、思考模式和并发数量。

The built-in defaults are designed for a DeepSeek-compatible chat-completions endpoint, but another compatible service can be used. 内置默认值面向兼容 DeepSeek 的聊天补全接口，也可以改用其他兼容服务。

Use AI翻译 to translate missing English tags on the current page into concise Chinese. 使用“AI翻译”可将当前页面缺少中文释义的英文标签翻译为简洁中文。

Use AI classification for the current prompt tab or run batch classification for all eligible character tabs. 可以对当前提示词标签页执行 AI 分类，也可以批量分类所有符合条件的人物标签页。

The API key and related settings are stored in the browser local storage, so manage them according to your own security requirements. API Key 和相关设置保存在浏览器本地存储中，请根据自身安全要求妥善管理。

Saving and Backups / 保存与备份

Click 保存（覆盖） to write changes back to the currently opened JSON file. 点击“保存（覆盖）”可将修改写回当前打开的 JSON 文件。

Click 另存为 to save a separate JSON copy and continue editing that new file. 点击“另存为”可保存独立的 JSON 副本，并继续编辑新文件。

The page creates one automatic browser-local backup every three hours and creates a deletion backup before destructive operations. 网页每三小时生成一份浏览器本地自动备份，并在删除操作前生成删除备份。

Backups can be downloaded or restored from Settings, and the number of retained deletion backups is configurable. 可在设置中下载或还原备份，并可配置保留的删除备份数量。

The orphan-dictionary cleaner finds translations whose English tags are no longer used and creates a backup before deleting them. 孤立词典清理功能会查找已无对应英文标签使用记录的翻译条目，并在删除前自动备份。

Usage Agreement / 使用协议

Read 使用协议.txt before using, redistributing, merging, continuing training, or commercially using the model or generated images. 在使用、转载、融模、继续训练或商业使用模型及生成图片前，请先阅读 使用协议.txt。

The agreement contains different attribution and commercial-use requirements for Soul Tide characters, xiaobai, and xiaotiannan. 协议对《灵魂潮汐》角色、xiaobai 与 xiaotiannan 分别规定了不同的署名和商业使用要求。

Users are responsible for complying with applicable laws, third-party rights, the Anima model license, and the included usage agreement. 使用者须自行确保遵守适用法律、第三方权利、Anima 模型协议以及仓库内的使用协议。