Initial Clean State

.gitattributes

@@ -0,0 +1,35 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,3 @@
+*cuda*
+my_model_bge_m3/
+*.ignore

Dockerfile

@@ -0,0 +1,11 @@
+FROM python:3.12
+WORKDIR /app
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+RUN useradd -m -u 1000 user
+USER user
+ENV HOME=/home/user \
+    PATH=/home/user/.local/bin:$PATH
+COPY --chown=user . /app
+EXPOSE 7860
+CMD ["python", "main.py"]

README.md

@@ -0,0 +1,112 @@
+---
+title: DanbooruSearch
+emoji: 👁
+colorFrom: red
+colorTo: indigo
+sdk: docker
+pinned: false
+license: mit
+short_description: 使用模糊搜索查找danbooru标签
+---
+
+## Danbooru模糊查找
+
+本项目提供一种模糊查找Danbooru标签的方法。用户可以凭借模糊的语言描述，查找Danbooru数据集中对应的标签。
+
+目前，仅支持使用汉语和英语进行查找。
+
+⚠查找结果中可能包含NSFW内容。
+
+### 使用说明
+
+本程序有三个查找参数：
+
+- `top_k`：每个词取前`top_k`个相关的标签。
+
+- `结果上限`：对于整个查找，展示综合得分的前N名
+
+- `热度权重`：0~1，这个数值越高，标签频数在综合得分中的权重越大。推荐设为0.15
+
+设置完查找参数后，用户可以在「画面描述」中输入汉语或英语的词语或自然语言，点击「开始搜索」后，程序将进行查找。
+
+### 工作模式
+
+#### 完整画面查找
+
+你可以输入对画面的完整描述，程序将生成匹配的danbooru标签集。
+
+**示例输入**
+
+```
+一个穿着白色水手服、蓝色短裙的少女，在下着大雨的城市街道奔跑，她的标签是不甘、愤怒、流泪，她的衣服湿透。
+```
+
+**建议参数**
+
+- `top_k`：较小，例如5
+
+- `结果上限`：一般的SDXL系生图模型支持70-80个标签，设为80
+
+- `热度权重`：0.15
+
+**示例输出**
+
+```
+short_dress, city, streaming_tears, street, white_serafuku, tag, hashtag, clothes, furious, fume, label, rain, blue_background, tears, miniskirt, white_sailor_collar, shorts_under_skirt, dressing, serafuku, costume, shorts_under_dress, dress, crying, blue_skin, undressing, microskirt, object_in_clothes, tearing_up, blue_theme, black_serafuku, running, blue_shoes, >:(, shared_clothes, clothes_on_and_off, blue_liquid, jogging, clothes_in_front, wiping_tears, grey_serafuku, urban, angry, shorts, fleeing, town, wet_clothes, la_grondement_du_haine, wet_face, slug_girl, price_tag, changing_clothes, clothes_on_floor, angry_sex, wet_shirt, training_bra, wet_hair, sword_girls, cityscape, annoyed, running_on_liquid, outdoors, short_sleeves, supergirl, teardrop, girly_running, after_rain, short_shorts, blue_eyes, taking_shelter, city_lights, |_|, running_towards_viewer, nipple_tag, wet_floor, showering, mourning, ^^^, wet_jacket, w, chasing
+```
+
+同时，程序输出一份详细的查询结果表格，用户也可以在表格中选择合适的标签
+
+![image-20260208212616475](https://akizukipic.oss-cn-beijing.aliyuncs.com/img/202602082126523.png)
+
+#### 关键词精准查找
+
+你或许对某个词有模糊的印象，但不知道它对应的danbooru标签具体是什么。此时，你可以使用此程序精确查找。
+
+**示例输入**
+
+```
+假肢
+```
+
+**建议参数**
+
+- `top_k`：较小，例如5
+
+- `结果上限`：较小，例如10
+
+- `热度权重`：0.15
+
+**示例输出**
+
+```
+running_blades, prosthetic_leg, peg_leg, severed_limb, fake_claws, fake_nails, detached_legs, multiple_legs, separated_legs, alternate_footwear
+```
+
+![image-20260208212925262](https://akizukipic.oss-cn-beijing.aliyuncs.com/img/202602082129286.png)
+
+#### 概念模糊查找
+
+你可能像搜索关于某一个概念的关键词。此时，你可以使用此程序进行概念模糊查找。
+
+**示例输入**
+
+```
+裙子
+```
+
+**建议参数**
+
+- `top_k`：较大，例如40
+
+- `结果上限`：较大，例如80
+
+- `热度权重`：0.15
+
+**示例输出**
+
+```
+skirt, dress, one_piece, white_skirt, black_skirt, upskirt, skirt_set, long_skirt, long_dress, pink_skirt, skirt_hold, skirt_lift, sweater_dress, skirt_suit, multicolored_skirt, skirt_pull, under_skirt, gown, dress_lift, unworn_skirt, skirt_tug, print_skirt, dress_tug, dress_flower, orange_skirt, flower_skirt, flower_dress, wet_dress, skirt_under_dress, wet_skirt, dress_ribbon, taut_skirt, armored_skirt, culottes, swimsuit_skirt, skort, skirt_rolled_up, dress_flip, skousers, dress_over_shirt, striped_clothes, dress_shirt, sleeveless_dress, undressing, clothes, lingerie, striped_dress, layered_dress, adjusting_clothes, multicolored_dress, clothes_between_thighs, hooded_dress, dressing, layered_clothes, clothes_on_floor, evening_gown, dress_shoes, hand_under_clothes, tearing_clothes, multicolored_clothes, clothed_female_nude_female, clothes_down, ballerina, holding_cloth, changing_clothes, shorts_under_dress, cloth, shirt_under_dress, clothes_on_and_off, folded_clothes, hand_under_dress, clothes_on_shoulders, dress_suit, reflective_clothes, night_clothes
+```
+
+![image-20260208213214790](https://akizukipic.oss-cn-beijing.aliyuncs.com/img/202602082132644.png)

danbooru_vectors_multiview_cpu_fp16.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f6e8c5a1bbdf44b6883927d6ab8c9c51f362eec806f84f5ac2686df49febf3d
+size 395689916

main.py

@@ -0,0 +1,701 @@
+import pandas as pd
+from sentence_transformers import SentenceTransformer, util
+import os
+import pickle
+import jieba
+import numpy as np
+from nicegui import ui, app, run
+import asyncio
+import torch
+import re
+import time
+
+
+def is_running_on_huggingface_space():
+    return os.environ.get("SPACE_ID") is not None
+
+
+# 核心逻辑类 (单例模式)
+class DanbooruTagger:
+    _instance = None
+    _lock = asyncio.Lock()
+
+    @classmethod
+    async def get_instance(cls):
+        """获取单例，如果未初始化则等待初始化"""
+        async with cls._lock:
+            if cls._instance is None:
+                cls._instance = cls()
+                await run.io_bound(cls._instance.load)
+            return cls._instance
+
+    def __init__(self, model_path=None, csv_file='tags_enhanced.csv',
+                 cache_file='danbooru_vectors_multiview_'):
+
+        local_model_path = 'my_model_bge_m3'
+        hf_model_id = 'BAAI/bge-m3'
+
+        if model_path:
+            self.model_path = model_path
+        elif os.path.exists(local_model_path):
+            print(f"DTOOL: 检测到本地模型文件夹 '{local_model_path}'，将使用本地模型。")
+            self.model_path = local_model_path
+        else:
+            print(f"DTOOL: 未找到本地模型，将使用 HuggingFace ID '{hf_model_id}'。")
+            self.model_path = hf_model_id
+
+        self.csv_path = csv_file
+        self.device = 'cpu' if torch.cuda.is_available() else 'cpu'
+        # [修改] 更改缓存文件名后缀，避免读取旧缓存报错，触发重新构建
+        self.cache_path = cache_file + self.device + '_fp16.pkl'
+
+        self.model = None
+        self.df = None
+        self.emb_en = None
+        self.emb_cn = None
+        self.emb_wiki = None
+        self.emb_cn_core = None  # [新增] 核心词向量
+
+        self.max_log_count = 15.0
+        self.is_loaded = False
+
+        # 停用词表
+        self.stop_words = {
+            ',', '.', ':', ';', '?', '!', '"', "'", '`',
+            '(', ')', '[', ']', '{', '}', '<', '>',
+            '-', '_', '=', '+', '/', '\\', '|', '@', '#', '$', '%', '^', '&', '*', '~',
+            '，', '。', '：', '；', '？', '！', '“', '”', '‘', '’',
+            '（', '）', '【', '】', '《', '》', '、', '…', '—', '·',
+            ' ', '\t', '\n', '\r',
+            '的', '地', '得', '了', '着', '过',
+            '是', '为', '被', '给', '把', '让', '由',
+            '在', '从', '自', '向', '往', '对', '于',
+            '和', '与', '及', '或', '且', '而', '但', '并', '即', '又', '也',
+            '啊', '吗', '吧', '呢', '噢', '哦', '哈', '呀', '哇',
+            '我', '你', '他', '她', '它', '我们', '你们', '他们',
+            '这', '那', '此', '其', '谁', '啥', '某', '每',
+            '这个', '那个', '这些', '那些', '这里', '那里',
+            '个', '位', '只', '条', '张', '幅', '件', '套', '双', '对', '副',
+            '种', '类', '群', '些', '点', '份', '部', '名',
+            '很', '太', '更', '最', '挺', '特', '好', '真',
+            '一', '一个', '一种', '一下', '一点', '一些',
+            '有', '无', '非', '没', '不'
+        }
+
+        # 类型映射表
+        self.cat_map = {
+            '0': 'General', '1': 'Artist', '3': 'Copyright', '4': 'Character', '5': 'Meta'
+        }
+
+    def load(self):
+        if self.is_loaded: return
+        t0 = time.time()
+
+        # 加载缓存
+        if not os.path.exists(self.cache_path):
+            print("\n" + "=" * 50)
+            print("DTOOL: 未找到四视图索引缓存，开始首次构建（这可能需要 1~3 分钟）...")
+            print("=" * 50 + "\n")
+            self._build_from_csv()
+        else:
+            print(f"DTOOL: 正在加载缓存索引 ({self.cache_path})...")
+            self._load_from_cache()
+
+            if os.path.exists(self.csv_path):
+                self._smart_update()
+
+        # 模型加载
+        if self.model is None:
+            self._load_model()
+
+        self._setup_jieba_from_memory()
+        self.is_loaded = True
+        print(f"DTOOL: 系统初始化完成，耗时 {time.time() - t0:.2f} 秒")
+
+    def _load_model(self):
+        print(f"DTOOL: 正在加载模型 (路径: {self.model_path}, Device: {self.device})...")
+        try:
+            self.model = SentenceTransformer(self.model_path, device=self.device)
+        except Exception as e:
+            print(f"DTOOL: 本地模型加载失败，尝试从 HF 拉取: {e}")
+            self.model = SentenceTransformer('BAAI/bge-m3', device=self.device)
+
+    def _read_csv_robust(self, path):
+        for enc in ['utf-8', 'gbk', 'gb18030']:
+            try:
+                return pd.read_csv(path, dtype=str, encoding=enc).fillna("")
+            except UnicodeDecodeError:
+                continue
+        raise ValueError("CSV 读取失败，请检查编码")
+
+    def _preprocess_raw_df(self, df):
+        df.dropna(subset=['name'], inplace=True)
+        df = df[df['name'].str.strip() != '']
+
+        for col in ['cn_name', 'category', 'wiki', 'nsfw']:
+            if col not in df.columns: df[col] = ''
+
+        df['category'] = df['category'].fillna('0')
+        df['nsfw'] = df['nsfw'].fillna('0')
+
+        for char in ['，', '|', '、']:
+            df['cn_name'] = df['cn_name'].str.replace(char, ',', regex=False)
+
+        if 'post_count' not in df.columns: df['post_count'] = 0
+        df['post_count'] = pd.to_numeric(df['post_count'], errors='coerce').fillna(0)
+
+        df['cn_name'] = df['cn_name'].fillna("")
+        df['wiki'] = df['wiki'].fillna("")
+
+        # [新增] 提取 cn_name 的第一个词作为核心词
+        # 逻辑：按逗号分割，取第一个，去除首尾空格
+        df['cn_core'] = df['cn_name'].str.split(',', n=1).str[0].str.strip()
+        df['cn_core'] = df['cn_core'].fillna("")
+
+        df.drop_duplicates(subset=['name'], inplace=True)
+        df.reset_index(drop=True, inplace=True)
+        return df
+
+    def _build_from_csv(self):
+        print(f"DTOOL: 正在全量读取 {self.csv_path} ...")
+        raw_df = self._read_csv_robust(self.csv_path)
+        print("DTOOL: 正在预处理数据...")
+        self.df = self._preprocess_raw_df(raw_df)
+        self.max_log_count = np.log1p(self.df['post_count'].max())
+        self._load_model()
+        self._encode_and_save()
+
+    def _encode_and_save(self):
+        print("DTOOL: 正在生成向量索引...")
+
+        print("DTOOL: 正在生成英文索引...")
+        self.emb_en = self.model.encode(self.df['name'].tolist(), batch_size=64, show_progress_bar=True,
+                                        convert_to_tensor=True)
+
+        print("DTOOL: 正在生成中文扩展索引...")
+        self.emb_cn = self.model.encode(self.df['cn_name'].tolist(), batch_size=64, show_progress_bar=True,
+                                        convert_to_tensor=True)
+
+        print("DTOOL: 正在生成释义索引...")
+        self.emb_wiki = self.model.encode(self.df['wiki'].tolist(), batch_size=64, show_progress_bar=True,
+                                          convert_to_tensor=True)
+
+        print("DTOOL: 正在生成中文核心词索引...")
+        self.emb_cn_core = self.model.encode(self.df['cn_core'].tolist(), batch_size=64, show_progress_bar=True,
+                                             convert_to_tensor=True)
+
+        print("DTOOL: 正在保存 (FP16)...")
+        cache_data = {
+            'df': self.df,
+            'embeddings_en': self.emb_en.half(),
+            'embeddings_cn': self.emb_cn.half(),
+            'embeddings_wiki': self.emb_wiki.half(),
+            'embeddings_cn_core': self.emb_cn_core.half(),  # [新增]
+            'max_log_count': self.max_log_count
+        }
+        with open(self.cache_path, 'wb') as f:
+            pickle.dump(cache_data, f)
+        print("DTOOL: 缓存保存成功！")
+
+    def _load_from_cache(self):
+        with open(self.cache_path, 'rb') as f:
+            data = pickle.load(f)
+            self.df = data['df']
+            self.emb_en = data['embeddings_en'].float()
+            self.emb_cn = data['embeddings_cn'].float()
+            self.emb_wiki = data.get('embeddings_wiki', torch.zeros_like(self.emb_en)).float()
+            self.emb_cn_core = data.get('embeddings_cn_core', torch.zeros_like(self.emb_en)).float()
+            self.max_log_count = data.get('max_log_count', 15.0)
+
+    def _smart_update(self):
+        print("DTOOL: 正在检查数据变更...")
+        raw_df = self._read_csv_robust(self.csv_path)
+        compare_df = raw_df.copy()
+
+        for col in ['cn_name', 'wiki', 'nsfw']:
+            if col not in compare_df.columns: compare_df[col] = ""
+            compare_df[col] = compare_df[col].fillna("")
+
+        for char in ['，', '|', '、']:
+            compare_df['cn_name'] = compare_df['cn_name'].str.replace(char, ',', regex=False)
+
+        current_map = {}
+        for _, row in compare_df.iterrows():
+            current_map[row['name']] = (row['cn_name'], row['wiki'], row['nsfw'])
+
+        cached_df = self.df
+        cached_map = {}
+        has_wiki = 'wiki' in cached_df.columns
+        has_nsfw = 'nsfw' in cached_df.columns
+        for _, row in cached_df.iterrows():
+            w = row['wiki'] if has_wiki else ""
+            n = row['nsfw'] if has_nsfw else "0"
+            cached_map[row['name']] = (row['cn_name'], w, n)
+
+        new_tags = []
+        changed_tags = []
+
+        for name, new_tuple in current_map.items():
+            if name not in cached_map:
+                new_tags.append(name)
+            else:
+                if new_tuple != cached_map[name]:
+                    changed_tags.append(name)
+
+        if not new_tags and not changed_tags:
+            print("DTOOL: 数据已是最新。")
+            return
+
+        print(f"DTOOL: 检测到变更 -> 新增: {len(new_tags)}, 修改: {len(changed_tags)}")
+        print("DTOOL: 检测到数据变���，触发重建索引...")
+        self._build_from_csv()
+
+    def _setup_jieba_from_memory(self):
+        print("DTOOL: 正在从内存构建 Jieba 词典...")
+        if self.df is not None:
+            target_col = 'cn_name'
+            texts = self.df[target_col].dropna().astype(str).tolist()
+            unique_words = set()
+            for text in texts:
+                parts = text.replace(',', ' ').split()
+                for part in parts:
+                    part = part.strip()
+                    if len(part) > 1: unique_words.add(part)
+            for word in unique_words:
+                jieba.add_word(word, 2000)
+
+    def _smart_split(self, text):
+        tokens = []
+        chunks = re.split(r'([\u4e00-\u9fa5]+)', text)
+        for chunk in chunks:
+            if not chunk.strip(): continue
+            if re.match(r'[\u4e00-\u9fa5]+', chunk):
+                tokens.extend(jieba.cut(chunk))
+            else:
+                cleaned = re.sub(r'[,()\[\]{}:]', ' ', chunk)
+                parts = cleaned.split()
+                for part in parts:
+                    try:
+                        float(part)
+                    except ValueError:
+                        tokens.append(part)
+        return tokens
+
+    def search(self, user_query, top_k=5, limit=80, popularity_weight=0.15, show_nsfw=False,
+               use_segmentation=True, target_layers=None, target_categories=None):
+        if not self.is_loaded: self.load()
+
+        if target_layers is None: target_layers = ['英文', '中文扩展词', '释义', '中文核心词']
+        if target_categories is None: target_categories = ['General', 'Character', 'Copyright']
+
+        if use_segmentation:
+            raw_keywords = self._smart_split(user_query)
+            keywords = [w.strip() for w in raw_keywords if w.strip() and w.strip() not in self.stop_words]
+            search_queries = [user_query] + keywords
+        else:
+            keywords = []
+            search_queries = [user_query]
+
+        query_embeddings = self.model.encode(search_queries, convert_to_tensor=True).float()
+
+        empty_hits = [[] for _ in search_queries]
+        hits_en = util.semantic_search(query_embeddings, self.emb_en,
+                                       top_k=top_k) if '英文' in target_layers else empty_hits
+        hits_cn = util.semantic_search(query_embeddings, self.emb_cn,
+                                       top_k=top_k) if '中文扩展词' in target_layers else empty_hits
+        hits_wiki = util.semantic_search(query_embeddings, self.emb_wiki,
+                                         top_k=top_k) if '释义' in target_layers else empty_hits
+        hits_cn_core = util.semantic_search(query_embeddings, self.emb_cn_core,
+                                            top_k=top_k) if '中文核心词' in target_layers else empty_hits
+
+        final_results = {}
+
+        for i, _ in enumerate(search_queries):
+            source_word = search_queries[i]
+            combined = []
+            for h in hits_en[i]: combined.append((h, '英文'))
+            for h in hits_cn[i]: combined.append((h, '中文扩展词'))
+            for h in hits_wiki[i]: combined.append((h, '释义'))
+            for h in hits_cn_core[i]: combined.append((h, '中文核心词'))
+
+            for hit, layer in combined:
+                score = hit['score']
+                if score < 0.35: continue
+
+                idx = hit['corpus_id']
+                row = self.df.iloc[idx]
+
+                nsfw_flag = str(row.get('nsfw', '0'))
+                if not show_nsfw and nsfw_flag == '1': continue
+
+                cat_val = str(row.get('category', '0'))
+                cat_text = self.cat_map.get(cat_val, 'Other')
+                if cat_text not in target_categories: continue
+
+                tag_name = row['name']
+                count = row['post_count']
+
+                log_count = np.log1p(count)
+                pop_score = log_count / self.max_log_count
+                final_score = (score * (1 - popularity_weight)) + (pop_score * popularity_weight)
+
+                if tag_name not in final_results or final_score > final_results[tag_name]['final_score']:
+                    final_results[tag_name] = {
+                        'tag': tag_name,
+                        'final_score': round(float(final_score), 4),
+                        'semantic_score': round(float(score), 4),
+                        'cn_name': row['cn_name'],
+                        'count': int(count),
+                        'source': source_word,
+                        'layer': layer,
+                        'category': cat_text,
+                        'nsfw': nsfw_flag,
+                        'wiki': str(row.get('wiki', ''))
+                    }
+
+        sorted_tags = sorted(final_results.items(), key=lambda x: x[1]['final_score'], reverse=True)
+        valid_tags = [item[1] for item in sorted_tags if item[1]['final_score'] > 0.45]
+
+        if len(valid_tags) > limit:
+            valid_tags = valid_tags[:limit]
+
+        tag_string = ", ".join([item['tag'] for item in valid_tags])
+
+        return tag_string, valid_tags, keywords
+
+
+# NiceGUI 界面
+
+# 定义基础列
+base_columns = [
+    {'name': 'tag', 'label': '匹配标签', 'field': 'tag', 'align': 'left', 'sortable': True},
+    {'name': 'cn_name', 'label': '含义', 'field': 'cn_name', 'align': 'left'},
+    {'name': 'category', 'label': '类型', 'field': 'category', 'align': 'left', 'sortable': True},
+    {'name': 'nsfw', 'label': '分级', 'field': 'nsfw', 'align': 'center', 'sortable': True},
+    {'name': 'final_score', 'label': '综合分', 'field': 'final_score', 'sortable': True},
+    {'name': 'count', 'label': '热度', 'field': 'count', 'sortable': True},
+]
+
+# 定义可选列
+optional_col_map = {
+    'semantic': {'name': 'semantic_score', 'label': '语义分', 'field': 'semantic_score', 'sortable': True},
+    'layer': {'name': 'layer', 'label': '匹配层', 'field': 'layer'},
+    'source': {'name': 'source', 'label': '匹配来源', 'field': 'source'},
+}
+
+
+@ui.page('/')
+async def main_page():
+    ui.colors(primary='#4A90E2', secondary='#5E6C84', accent='#FF6B6B')
+
+    full_table_data = []
+    current_query_str = ""
+
+    # 提示区
+    with ui.card().classes('w-full max-w-6xl mx-auto bg-orange-50 border-l-4 border-orange-500 mb-2'):
+        with ui.column().classes('gap-1'):
+            ui.label('⚠️ 注意事项 / Note').classes('text-lg font-bold text-orange-800')
+            ui.markdown("""
+                        - **本网站为AI工具，其结果未必正确无误** (Results may contain errors)
+                        - **查找结果可能会包括 NSFW 内容** (Results may include NSFW content)
+                        - **仅支持汉语、英语查找** (Only supports Chinese/English)
+                        - **仅显示Danbooru频数超过100的标签** (Frequency >= 100)
+                        - **仅显示特征、角色、作品标签**  (General Character and Copyright tags only)
+                        - **Comfy UI 插件地址** ： https://github.com/SuzumiyaAkizuki/ComfyUI-DanbooruSearcher
+                        """).classes('text-sm text-gray-800 ml-4')
+
+    # 主体
+    with ui.column().classes('w-full max-w-6xl mx-auto p-4 gap-6'):
+        with ui.row().classes('items-center gap-2'):
+            ui.icon('search', size='2em', color='primary')
+            ui.label('Danbooru 标签模糊搜索').classes('text-2xl font-bold text-gray-800')
+
+        # 基础控制面板
+        with ui.card().classes('w-full'):
+            with ui.grid(columns=4).classes('w-full gap-8 items-center'):
+                input_top_k = ui.number('Top K (语义相关)', value=5, min=1, max=50) \
+                    .props('outlined dense suffix="个"').classes('w-full')
+
+                input_limit = ui.number('结果上限', value=80, min=10, max=500) \
+                    .props('outlined dense suffix="个"').classes('w-full')
+
+                with ui.column().classes('gap-0'):
+                    with ui.row().classes('w-full justify-between'):
+                        ui.label('热度权重').classes('text-xs text-gray-500')
+                        ui.label().bind_text_from(input_weight := ui.slider(min=0.0, max=1.0, value=0.15, step=0.05),
+                                                  'value', lambda v: f"{v:.2f}")
+                    input_weight.classes('w-full')
+
+                input_nsfw = ui.switch('显示 NSFW', value=False).props('color=red').classes('w-full')
+
+        # 高级设置面板
+        with ui.expansion('高级设置 (Advanced Settings)', icon='tune').classes('w-full bg-gray-50 border rounded-lg'):
+            with ui.column().classes('w-full p-4 gap-4'):
+
+                # 分词开关
+                input_segment = ui.switch('启用智能分词 (Segmentation)', value=True).props('color=primary')
+                ui.label('关闭后系统将只匹配完整句子，适用于精准搜索整句。').classes('text-xs text-gray-500 -mt-2 ml-10')
+
+                ui.separator()
+
+                # 匹配层筛选
+                ui.label('匹配层筛选 (Target Layers):').classes('font-bold text-gray-700')
+                with ui.row().classes('gap-4'):
+                    layer_options = ['英文', '中文扩展词', '释义', '中文核心词']
+                    selected_layers = {layer: True for layer in layer_options}
+
+                    def toggle_layer(l, value): selected_layers[l] = value
+
+                    for layer in layer_options:
+                        ui.checkbox(layer, value=True, on_change=lambda e, l=layer: toggle_layer(l, e.value))
+
+                ui.separator()
+
+                # 类型筛选
+                ui.label('标签类型筛选 (Categories):').classes('font-bold text-gray-700')
+                with ui.row().classes('gap-4 flex-wrap'):
+                    cat_options = ['General', 'Copyright', 'Character']
+                    selected_cats = {cat: True for cat in cat_options}
+
+                    def toggle_cat(c, value): selected_cats[c] = value
+
+                    for cat in cat_options:
+                        color_map = {'General': 'blue', 'Copyright': 'pink', 'Character': 'green'}
+                        ui.checkbox(cat, value=True, on_change=lambda e, c=cat: toggle_cat(c, e.value)) \
+                            .props(f'color={color_map.get(cat, "primary")}')
+
+                ui.separator()
+
+                # 表格显示选项
+                ui.label('表格显示选项 (Display Options):').classes('font-bold text-gray-700')
+                with ui.row().classes('gap-6'):
+                    sw_semantic = ui.switch('显示语义分', value=False)
+                    sw_layer = ui.switch('显示匹配层', value=False)
+                    sw_source = ui.switch('显示匹配来源', value=False)
+
+                    # 动态更新表格列
+                    def update_table_columns():
+                        cols = list(base_columns)  # 复制基础列
+                        if sw_semantic.value: cols.append(optional_col_map['semantic'])
+                        if sw_layer.value: cols.append(optional_col_map['layer'])
+                        if sw_source.value: cols.append(optional_col_map['source'])
+                        result_table.columns = cols
+
+                    sw_semantic.on('update:model-value', update_table_columns)
+                    sw_layer.on('update:model-value', update_table_columns)
+                    sw_source.on('update:model-value', update_table_columns)
+
+
+        # 搜索输入区
+        with ui.card().classes('w-full p-0 overflow-hidden'):
+            with ui.column().classes('w-full p-6 gap-4'):
+                ui.label('画面描述').classes('text-lg font-bold text-gray-700')
+                search_input = ui.textarea(placeholder='例如：一个穿着白色水手服的女孩在雨中奔跑').classes(
+                    'w-full text-lg').props('outlined rows=3')
+
+                keywords_container = ui.row().classes('gap-2 items-center')
+                spinner = ui.spinner(size='2em').classes('hidden')
+
+                def filter_table_by_source(keyword):
+                    if not keyword or keyword == 'ALL':
+                        result_table.rows = full_table_data
+                    else:
+                        result_table.rows = [row for row in full_table_data if row['source'] == keyword]
+
+                    for child in keywords_container.default_slot.children:
+                        if isinstance(child, ui.chip):
+                            is_selected = False
+                            if keyword == 'ALL' and child.text == '全部':
+                                is_selected = True
+                            elif keyword == current_query_str and child.text == '整句':
+                                is_selected = True
+                            elif child.text == keyword:
+                                is_selected = True
+
+                            child.props(
+                                f'color={"primary" if is_selected else "grey-4"} text-color={"white" if is_selected else "black"}')
+
+                async def perform_search():
+                    nonlocal full_table_data, current_query_str
+                    query = search_input.value.strip()
+                    if not query: return
+
+                    current_query_str = query
+                    search_btn.disable()
+                    spinner.classes(remove='hidden')
+                    ui.notify('正在搜索...', type='info')
+
+                    target_layers_list = [k for k, v in selected_layers.items() if v]
+                    target_cats_list = [k for k, v in selected_cats.items() if v]
+
+                    if not target_layers_list:
+                        ui.notify('请至少选择一个匹配层！', type='warning')
+                        search_btn.enable()
+                        spinner.classes(add='hidden')
+                        return
+
+                    try:
+                        tagger = await DanbooruTagger.get_instance()
+
+                        tags_str, table_data, keywords = await run.io_bound(
+                            tagger.search,
+                            query,
+                            int(input_top_k.value),
+                            int(input_limit.value),
+                            float(input_weight.value),
+                            input_nsfw.value,
+                            input_segment.value,
+                            target_layers_list,
+                            target_cats_list
+                        )
+
+                        full_table_data = table_data
+
+                        all_result_area.value = tags_str
+                        result_table.rows = table_data
+                        result_table.selected = []
+                        update_selection_display(None)
+
+                        keywords_container.clear()
+                        with keywords_container:
+                            ui.label('分词筛选:').classes('text-sm text-gray-500 font-bold mr-2')
+                            ui.chip('全部', on_click=lambda: filter_table_by_source('ALL')) \
+                                .props('color=primary text-color=white clickable')
+
+                            if input_segment.value:
+                                ui.chip('整句', on_click=lambda: filter_table_by_source(current_query_str)) \
+                                    .props('color=grey-4 text-color=black clickable')
+                                for kw in keywords:
+                                    ui.chip(kw, on_click=lambda k=kw: filter_table_by_source(k)) \
+                                        .props('color=grey-4 text-color=black clickable')
+                            else:
+                                ui.label('(分词已关闭)').classes('text-xs text-gray-400')
+
+                        ui.notify(f'找到 {len(table_data)} 个标签', type='positive')
+
+                    except Exception as e:
+                        ui.notify(f'错误: {str(e)}', type='negative')
+                    finally:
+                        search_btn.enable()
+                        spinner.classes(add='hidden')
+
+                with ui.row().classes('w-full justify-end items-center gap-4'):
+                    spinner
+                    search_btn = ui.button('开始搜索', on_click=perform_search, icon='search')
+                    search_btn.classes('px-8 py-2 text-lg').props('unelevated color=primary')
+
+                search_input.on('keydown.ctrl.enter', perform_search)
+
+        # 结果表格
+        with ui.row().classes('w-full gap-6'):
+            with ui.card().classes('w-1/3 flex-grow'):
+                ui.label('推荐 Prompt (全部)').classes('font-bold text-gray-600')
+                all_result_area = ui.textarea().classes('w-full h-full bg-gray-50').props(
+                    'readonly outlined input-class=text-sm')
+
+            with ui.column().classes('w-2/3 flex-grow'):
+                with ui.card().classes('w-full bg-blue-50 border-blue-200 border'):
+                    with ui.row().classes('w-full items-center justify-between'):
+                        with ui.row().classes('items-center gap-2'):
+                            ui.icon('check_circle', color='primary')
+                            ui.label('已选标签:').classes('font-bold text-primary')
+                            selection_count_label = ui.label('0').classes(
+                                'bg-primary text-white px-2 rounded-full text-sm')
+                        copy_btn = ui.button('复制选中', icon='content_copy').props('dense unelevated color=primary')
+
+                    selected_display = ui.textarea().classes('w-full mt-2').props(
+                        'outlined dense rows=2 readonly bg-white')
+
+                    def copy_selection():
+                        ui.clipboard.write(selected_display.value)
+                        ui.notify('已复制选中标签!', type='positive')
+
+                    copy_btn.on_click(copy_selection)
+
+                def update_selection_display(e):
+                    selected_rows = result_table.selected
+                    tags = [row['tag'] for row in selected_rows]
+                    selected_display.value = ", ".join(tags)
+                    selection_count_label.text = str(len(tags))
+
+                result_table = ui.table(
+                    columns=base_columns,
+                    rows=[],
+                    pagination=10,
+                    selection='multiple',
+                    row_key='tag'
+                ).classes('w-full')
+
+                result_table.on('selection', update_selection_display)
+
+                result_table.add_slot('body-cell-final_score', '''
+                    <q-td :props="props">
+                        <q-badge :color="props.value > 0.6 ? 'green' : (props.value > 0.5 ? 'teal' : 'orange')">
+                            {{ props.value }}
+                        </q-badge>
+                    </q-td>
+                ''')
+
+                result_table.add_slot('body-cell-category', '''
+                    <q-td :props="props">
+                        <q-badge :color="
+                            props.value === 'General' ? 'blue' : 
+                            (props.value === 'Character' ? 'green' : 
+                            (props.value === 'Copyright' ? 'pink' : 'red'))
+                        " outline>
+                            {{ props.value }}
+                        </q-badge>
+                    </q-td>
+                ''')
+
+                result_table.add_slot('body-cell-nsfw', '''
+                    <q-td :props="props">
+                        <div v-if="props.value === '1'" title="NSFW Content" class="text-red-500">🔴</div>
+                        <div v-else class="text-green-500">🟢</div>
+                    </q-td>
+                ''')
+
+                # 4. 中文含义分词显示 + 悬停显示 Wiki (修复字体大小问题)
+                result_table.add_slot('body-cell-cn_name', '''
+                                    <q-td :props="props">
+                                        <template v-if="props.value">
+                                            <q-badge 
+                                                v-for="(item, index) in props.value.split(',')" 
+                                                :key="index"
+                                                :color="index === 0 ? 'black' : 'grey'"
+                                                outline 
+                                                style="font-size: 14px" 
+                                                class="q-mr-xs q-mb-xs cursor-help"
+                                            >
+                                                {{ item }}
+                                            </q-badge>
+                                        </template>
+                                        <q-tooltip 
+                                            v-if="props.row.wiki" 
+                                            content-class="bg-black text-white shadow-4" 
+                                            max-width="500px"
+                                            :offset="[10, 10]"
+                                        >
+                                            <div style="font-size: 14px; line-height: 1.5;">
+                                                {{ props.row.wiki }}
+                                            </div>
+                                        </q-tooltip>
+                                    </q-td>
+                                ''')
+
+
+if __name__ in {"__main__", "__mp_main__"}:
+    if is_running_on_huggingface_space():
+        host = '0.0.0.0'
+        port = 7860
+    else:
+        host = '127.0.0.1'
+        port = 1145
+
+    ui.run(
+        host=host,
+        port=port,
+        title='Danbooru Tags Searcher',
+        reload=True,
+        show=True
+    )

requirements.txt

@@ -0,0 +1,5 @@
+nicegui
+pandas
+sentence-transformers
+torch
+jieba