diff --git a/core/backend/Dockerfile b/core/backend/Dockerfile
index bb99c8a..64dc243 100644
--- a/core/backend/Dockerfile
+++ b/core/backend/Dockerfile
@@ -32,6 +32,8 @@ RUN chmod +x /usr/local/bin/docker-entrypoint.sh
 
 EXPOSE 8000
 ENTRYPOINT ["docker-entrypoint.sh"]
-# timeout 300:前端已把多图生成拆成「每张一条请求」,单张约 30s;300s 给慢 ARK 留充足余量,
-# 不再因一条请求串行出多张而超时(旧 120s 下生成 4 张必被 worker 强杀 → 500)。
-CMD ["gunicorn", "airshelf.wsgi:application", "--bind", "0.0.0.0:8000", "--workers", "3", "--timeout", "300"]
+# timeout 300:图片生成是慢请求(等 ARK+传 TOS ~30s),300s 给慢 ARK 留充足余量。
+# --threads 4:关键!纯 sync worker 下,几张图就能把 3 个 worker 占满,健康探针 /api/health/
+# 拿不到 worker → 超时 → pod 被判不就绪 → 流量被掐 → 整站 502(表现:报错但后端在干活、刷新能看到图)。
+# 图片生成是 I/O 密集(等 ARK/TOS 时线程让出 GIL),3 worker × 4 线程 = 12 并发,慢请求不再饿死探针。
+CMD ["gunicorn", "airshelf.wsgi:application", "--bind", "0.0.0.0:8000", "--workers", "3", "--threads", "4", "--timeout", "300"]
diff --git a/core/frontend/src/App.tsx b/core/frontend/src/App.tsx
index 6b0049f..2b3e133 100644
--- a/core/frontend/src/App.tsx
+++ b/core/frontend/src/App.tsx
@@ -355,17 +355,28 @@ export function App() {
   }
 
   function generateImages(payload: { prompt: string; mode?: "image" | "model" | "cover"; count?: number }) {
-    // 每张图一条独立 HTTP 请求、并行发出(不再让一条请求串行出 N 张)——
-    // 单张约 30s 不会触发后端请求超时;并行后 N 张总耗时≈单张,也不长时间占用后端进程。
+    // 每张图一条独立 HTTP 请求(不再让一条请求串行出 N 张)——单张约 30s 不会触发后端请求超时。
+    // 但不全量并行:同时最多 CONCURRENCY 条,避免 N 条一起砸向单个后端 pod 致内存峰值翻倍 OOM→502。
     // 任一张失败只丢那一张(各自独立扣费/回滚),其余正常返回。
     const n = Math.max(1, Math.min(Number(payload.count) || 1, 12));
+    const CONCURRENCY = 2;
     return action(async () => {
-      const settled = await Promise.allSettled(
-        Array.from({ length: n }, () => api.generateImage({ prompt: payload.prompt, mode: payload.mode, count: 1 }))
-      );
-      const assets = settled.flatMap((r) => (r.status === "fulfilled" ? r.value.assets : []));
+      const results: Array<{ ok: true; assets: Asset[] } | { ok: false; reason: unknown }> = new Array(n);
+      let next = 0;
+      async function worker() {
+        for (let i = next++; i < n; i = next++) {
+          try {
+            const res = await api.generateImage({ prompt: payload.prompt, mode: payload.mode, count: 1 });
+            results[i] = { ok: true, assets: res.assets };
+          } catch (reason) {
+            results[i] = { ok: false, reason };
+          }
+        }
+      }
+      await Promise.all(Array.from({ length: Math.min(CONCURRENCY, n) }, worker));
+      const assets = results.flatMap((r) => (r.ok ? r.assets : []));
       if (assets.length === 0) {
-        const firstErr = settled.find((r) => r.status === "rejected") as PromiseRejectedResult | undefined;
+        const firstErr = results.find((r) => !r.ok) as { ok: false; reason: unknown } | undefined;
         throw new Error(firstErr ? String((firstErr.reason as Error)?.message || firstErr.reason || "生成失败") : "未生成任何图片");
       }
       return { assets };
diff --git a/core/frontend/src/routes/ai-tools.tsx b/core/frontend/src/routes/ai-tools.tsx
index 5d9fbf9..0cb0c0c 100644
--- a/core/frontend/src/routes/ai-tools.tsx
+++ b/core/frontend/src/routes/ai-tools.tsx
@@ -490,6 +490,8 @@ export function ImageWorkbenchPage({
   const [gridSort, setGridSort] = useState<"recent" | "name">("recent");
   const [tbOpen, setTbOpen] = useState<"" | "time" | "model">("");
   const [searchOpen, setSearchOpen] = useState(false);
+  // 模特网格默认只露 6 个;「全部模特 →」展开后露全部,再点收起(超过 6 个时第 7 个之后本来选不到)
+  const [showAllModels, setShowAllModels] = useState(false);
   function pickReference(event: ChangeEvent<HTMLInputElement>) {
     const file = event.target.files?.[0];
     if (!file) return;
@@ -834,14 +836,16 @@ export function ImageWorkbenchPage({
                   <div className="iw-step-h">
                     <span className="num">1</span>
                     <span className="title">选择模特</span>
-                    <span className="right">全部模特 →</span>
+                    {personAssets.length > 6 && (
+                      <span className="right" role="button" tabIndex={0} style={{ cursor: "pointer" }} onClick={() => setShowAllModels((v) => !v)} onKeyDown={(e) => { if (e.key === "Enter" || e.key === " ") { e.preventDefault(); setShowAllModels((v) => !v); } }}>{showAllModels ? "收起 ↑" : `全部模特 (${personAssets.length}) →`}</span>
+                    )}
                   </div>
                   <div className="model-grid">
                     {personAssets.length > 0
                       ? personAssets
                           .filter((item) => !gridQuery || item.name.toLowerCase().includes(gridQuery.toLowerCase()))
                           .sort((a, b) => (gridSort === "name" ? a.name.localeCompare(b.name) : (b.created_at || "").localeCompare(a.created_at || "")))
-                          .slice(0, 6)
+                          .slice(0, showAllModels ? undefined : 6)
                           .map((item) => {
                           const url = item.files?.[0]?.preview_url;
                           return (