feat: add /api/generate/poll endpoint for client-side task polling

New endpoint accepts POST with taskId/provider/modelId and returns
either a polling-continue response or the final media result. Each
request is short-lived (~1s) instead of holding a connection open.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/app/api/generate/poll/route.ts

@@ -0,0 +1,114 @@
+/**
+ * Poll API Route
+ *
+ * Handles status polling for long-running Kie.ai tasks.
+ * The client calls this endpoint repeatedly with short-lived requests
+ * instead of holding a single connection open for minutes.
+ */
+import { NextRequest, NextResponse } from "next/server";
+import type { GenerateResponse } from "@/types";
+import { checkKieTaskOnce, fetchKieMediaResult, isVeoModel } from "../providers/kie";
+import { buildMediaResponse } from "../route";
+
+export const maxDuration = 120; // 2 min — enough for media fetch, not for polling
+export const dynamic = 'force-dynamic';
+
+interface PollRequest {
+  taskId: string;
+  provider: string;
+  modelId: string;
+  modelName: string;
+  mediaType: string;
+}
+
+export async function POST(request: NextRequest) {
+  const requestId = Math.random().toString(36).substring(7);
+
+  try {
+    const body: PollRequest = await request.json();
+    const { taskId, provider, modelId, modelName, mediaType } = body;
+
+    if (!taskId || !provider) {
+      return NextResponse.json<GenerateResponse>(
+        { success: false, error: "taskId and provider are required" },
+        { status: 400 }
+      );
+    }
+
+    if (provider !== 'kie') {
+      return NextResponse.json<GenerateResponse>(
+        { success: false, error: `Unsupported poll provider: ${provider}` },
+        { status: 400 }
+      );
+    }
+
+    // Get API key (same pattern as route.ts)
+    const apiKey = request.headers.get("X-Kie-Key") || process.env.KIE_API_KEY;
+    if (!apiKey) {
+      return NextResponse.json<GenerateResponse>(
+        { success: false, error: "Kie.ai API key not configured" },
+        { status: 401 }
+      );
+    }
+
+    const isVeo = isVeoModel(modelId);
+    const pollResult = await checkKieTaskOnce(requestId, apiKey, taskId, isVeo);
+
+    if (pollResult.status === "processing") {
+      return NextResponse.json<GenerateResponse>({
+        success: true,
+        polling: true,
+        taskId,
+        pollProvider: provider,
+        pollModelId: modelId,
+        pollModelName: modelName,
+        pollMediaType: mediaType,
+      });
+    }
+
+    if (pollResult.status === "failed") {
+      return NextResponse.json<GenerateResponse>(
+        { success: false, error: `${modelName}: ${pollResult.error}` },
+        { status: 500 }
+      );
+    }
+
+    // completed — fetch media and return final result
+    const capabilityMap: Record<string, string[]> = {
+      video: ["text-to-video"],
+      audio: ["text-to-audio"],
+      image: ["text-to-image"],
+      "3d": ["text-to-3d"],
+    };
+
+    const result = await fetchKieMediaResult(requestId, {
+      pollData: pollResult.data!,
+      isVeo,
+      modelName,
+      capabilities: capabilityMap[mediaType] || ["text-to-image"],
+    });
+
+    if (!result.success) {
+      return NextResponse.json<GenerateResponse>(
+        { success: false, error: result.error || "Failed to fetch result" },
+        { status: 500 }
+      );
+    }
+
+    const output = result.outputs?.[0];
+    if (!output?.data && !output?.url) {
+      return NextResponse.json<GenerateResponse>(
+        { success: false, error: "No output in generation result" },
+        { status: 500 }
+      );
+    }
+
+    return buildMediaResponse(output);
+  } catch (error) {
+    console.error(`[API:${requestId}] Poll error:`, error);
+    return NextResponse.json<GenerateResponse>(
+      { success: false, error: error instanceof Error ? error.message : "Poll failed" },
+      { status: 500 }
+    );
+  }
+}