diff --git a/frontend/src/views/MediaList.vue b/frontend/src/views/MediaList.vue
index b02913f..4f9b3ba 100644
--- a/frontend/src/views/MediaList.vue
+++ b/frontend/src/views/MediaList.vue
@@ -118,6 +118,9 @@
                         <el-tag type="info" size="small">安全</el-tag>
                         <div class="nsfw-threshold">{{ formatAnimeRatings(scope.row.animeNsfwRatings) }}</div>
                     </div>
+                    <div v-else-if="scope.row.animeNsfwThreshold > 0">
+                        <el-tag type="info" size="small">安全</el-tag>
+                    </div>
                     <span v-else class="muted">-</span>
                 </template>
             </el-table-column>
diff --git a/internal/anime_nsfw/server.py b/internal/anime_nsfw/server.py
index a0609d6..38116be 100644
--- a/internal/anime_nsfw/server.py
+++ b/internal/anime_nsfw/server.py
@@ -272,16 +272,64 @@ class Handler(BaseHTTPRequestHandler):
         except Exception as exc:
             return self._send({"ok": False, "error": str(exc)})
 
-        # NSFW 判定：explicit 或 questionable 分数 >= 阈值，且必须高于 general。
-        # WD Tagger 的 rating 是独立 sigmoid（不归一化），非动漫图四个分值往往
-        # 都很接近；若 general 最高说明模型认为是安全图，只看绝对值会大量误判。
+        # NSFW 判定逻辑：
+        # WD Tagger v3 的 rating 是独立 sigmoid（不归一化），四个值可以同时很高。
+        # 对于某些动漫图，rating 没有区分度（都在 0.5-0.6 之间）。
+        #
+        # 判定策略：
+        #   1. 检查 top_tags 中是否有明确的色情标签（nude、genitals 等）
+        #   2. 如果有，判定为 NSFW
+        #   3. 否则，用 rating 辅助判断（explicit - general >= 0.1）
+        
         ratings = result["ratings"]
-        nsfw_raw = max(ratings["explicit"], ratings["questionable"])
-        if nsfw_raw > ratings["general"]:
-            nsfw_score = nsfw_raw
-            nsfw = nsfw_score >= threshold
+        top_tags = result["top_tags"]
+        explicit = ratings["explicit"]
+        questionable = ratings["questionable"]
+        general = ratings["general"]
+        
+        # 打印 top_tags 到日志（调试用）
+        if top_tags:
+            tag_str = ", ".join([f"{t['tag']}:{t['score']:.2f}" for t in top_tags[:5]])
+            print(f"=== AnimeNSFW 标签: {tag_str}", flush=True)
+        
+        # 定义明确的色情标签
+        NSFW_TAGS = {
+            "nude", "naked", "genitals", "penis", "vagina", "breasts",
+            "nipples", "areola", "pubic_hair", "anus", "testicles",
+            "erection", "ejaculation", "orgasm", "masturbation",
+            "fellatio", "cunnilingus", "intercourse", "sex",
+            "pussy", "cock", "dick", "boobs", "tits", "ass",
+            "butt", "thighs", "legs", "feet", "toes",
+        }
+        
+        # 检查 top_tags 中是否有明确的色情标签
+        nsfw_by_tags = False
+        nsfw_tag_score = 0.0
+        for tag_info in top_tags:
+            tag_name = tag_info["tag"].lower()
+            tag_score = tag_info["score"]
+            # 检查标签名是否包含色情关键词
+            for nsfw_keyword in NSFW_TAGS:
+                if nsfw_keyword in tag_name and tag_score > 0.5:
+                    nsfw_by_tags = True
+                    nsfw_tag_score = max(nsfw_tag_score, tag_score)
+                    break
+            if nsfw_by_tags:
+                break
+        
+        # 判定逻辑
+        # nsfw_score 始终返回 max(explicit, questionable) 的最高分，
+        # 这样 Go 侧可以通过 score > 0 判断检测器是否运行过。
+        # nsfw 标记才表示是否真正判定为色情。
+        nsfw_score = max(explicit, questionable)
+
+        if nsfw_by_tags:
+            # 标签检测到明确的色情内容
+            nsfw = True
+        elif nsfw_score >= threshold and (nsfw_score - general) >= 0.1:
+            # rating 辅助判断：最高分 >= 阈值 且 明显高于 general
+            nsfw = True
         else:
-            nsfw_score = 0.0
             nsfw = False
 
         self._send({
diff --git a/internal/organize/organize.go b/internal/organize/organize.go
index 274cfaa..7370621 100644
--- a/internal/organize/organize.go
+++ b/internal/organize/organize.go
@@ -609,7 +609,7 @@ func (o *Organizer) ProcessImages(images []scanner.File) {
 					}
 				}
 			}
-			if info != nil && r.animeNsfwScore > 0 {
+			if info != nil && (r.animeNsfwScore > 0 || r.animeNsfwRatings.General > 0) {
 				info.AnimeNsfwScore = r.animeNsfwScore
 				info.AnimeNsfwThreshold = o.animeNsfw.Threshold()
 				if ratingsJSON, err := json.Marshal(r.animeNsfwRatings); err == nil {