ai-chatbot/app/api/video-canvas/prompt-tools/route.ts

import { generateText } from "ai";
import { NextResponse } from "next/server";
import { z } from "zod";
import { auth } from "@/app/(auth)/auth";
import { getLanguageModelById } from "@/lib/ai/dynamic-provider";
import { getCompatibleTextGenerationTemperature } from "@/lib/ai/model-compat";
import {
  isImageGenerationModelDescriptor,
  isVideoModelDescriptor,
} from "@/lib/ai/model-classification";
import { getAiModelAgentCodingModelId } from "@/lib/db/ai-model-flags";
import {
  getAgentAiModelWithProvider,
  getEnabledAiModelsWithProvider,
  getSystemSetting,
} from "@/lib/db/queries";
import { createLogger } from "@/lib/observability/logger";
import {
  VIDEO_CANVAS_FUNCTION_CONFIG_KEY,
  defaultVideoCanvasFunctionConfig,
  normalizeVideoCanvasFunctionConfig,
} from "@/lib/video-canvas/config";
import {
  buildVideoPromptStructureInstructions,
  injectVideoPromptStructure,
  resolveVideoPromptStructure,
} from "@/lib/video-canvas/prompt-structure";

export const runtime = "nodejs";
export const maxDuration = 60;

const logger = createLogger("video-canvas-prompt-tools");

const requestSchema = z.object({
  marks: z
    .array(
      z.object({
        id: z.string().trim().min(1).max(100),
        label: z.string().trim().min(1).max(100),
      })
    )
    .max(12)
    .default([]),
  mode: z.enum(["optimize", "translate"]),
  prompt: z.string().trim().min(1, "请输入提示词").max(50_000, "提示词过长"),
  references: z
    .array(
      z.object({
        id: z.string().trim().min(1).max(100),
        index: z.number().int().min(1).max(10),
        label: z.string().trim().min(1).max(100),
        mediaType: z.enum(["audio", "image", "video"]),
        offset: z.number().int().min(0).max(50_000),
      })
    )
    .max(20)
    .default([]),
});

type ModelCandidate = Awaited<
  ReturnType<typeof getEnabledAiModelsWithProvider>
>[number];

function isTextModelCandidate(candidate: ModelCandidate) {
  const descriptor = {
    capabilities: candidate.model.capabilities,
    modelId: candidate.model.modelId,
    providerType: candidate.provider.type,
    supportsImageGen: candidate.model.supportsImageGen,
  };

  return (
    candidate.model.isEnabled &&
    candidate.provider.isEnabled &&
    !candidate.model.isSubModel &&
    candidate.model.healthStatus !== "auto_disabled" &&
    !isImageGenerationModelDescriptor(descriptor) &&
    !isVideoModelDescriptor(descriptor)
  );
}

function normalizeModelRef(value: unknown) {
  const normalized = String(value || "").trim();
  return normalized && normalized !== "auto" ? normalized : "";
}

function normalizeGeneratedText(value: unknown) {
  return String(value || "")
    .trim()
    .replace(/^```(?:text|markdown|md)?\s*/i, "")
    .replace(/\s*```$/i, "")
    .replace(/^(?:优化后的提示词|译文|translation)\s*[:：]\s*/i, "")
    .trim();
}

export async function POST(request: Request) {
  try {
    const session = await auth();
    if (!session?.user?.id) {
      return NextResponse.json({ success: false, error: "未登录" }, { status: 401 });
    }

    const parsed = requestSchema.safeParse(await request.json());
    if (!parsed.success) {
      return NextResponse.json(
        { success: false, error: parsed.error.issues[0]?.message || "请求参数无效" },
        { status: 400 },
      );
    }

    const setting = await getSystemSetting({ key: VIDEO_CANVAS_FUNCTION_CONFIG_KEY }).catch(
      () => null,
    );
    const config = setting?.value
      ? normalizeVideoCanvasFunctionConfig(setting.value)
      : defaultVideoCanvasFunctionConfig;
    const functionKey =
      parsed.data.mode === "optimize"
        ? "video-prompt-optimize"
        : "video-prompt-translate";
    const definition = config.functions[functionKey];

    if (!definition?.enabled) {
      return NextResponse.json(
        { success: false, error: `${definition?.label || "当前功能"}已停用` },
        { status: 403 },
      );
    }

    const [allModels, agentModel] = await Promise.all([
      getEnabledAiModelsWithProvider(false),
      getAgentAiModelWithProvider(),
    ]);
    const candidates = allModels.filter(isTextModelCandidate);
    const preferredModelRef =
      normalizeModelRef(definition.model) || normalizeModelRef(config.defaultMultimodalModel);
    const preferredModel = preferredModelRef
      ? candidates.find(
          (candidate) =>
            candidate.model.id === preferredModelRef ||
            candidate.model.modelId === preferredModelRef,
        )
      : null;
    const orderedCandidates = [
      preferredModel,
      agentModel && isTextModelCandidate(agentModel) ? agentModel : null,
      ...candidates,
    ].filter(
      (candidate, index, items): candidate is ModelCandidate =>
        Boolean(candidate) &&
        items.findIndex((item) => item?.model.id === candidate?.model.id) === index,
    );

    if (orderedCandidates.length === 0) {
      return NextResponse.json(
        { success: false, error: "当前站点没有可用的提示词处理模型" },
        { status: 500 },
      );
    }

    for (const candidate of orderedCandidates.slice(0, 3)) {
      try {
        const codingModelId = getAiModelAgentCodingModelId(candidate.model);
        const model = await getLanguageModelById(candidate.model.id, codingModelId);
        if (!model) continue;
        const structuredPrompt = injectVideoPromptStructure(
          parsed.data.prompt,
          parsed.data.references,
          parsed.data.marks,
        );
        const structureInstructions = buildVideoPromptStructureInstructions(
          parsed.data.references,
          parsed.data.marks,
        );

        const { text } = await generateText({
          abortSignal: AbortSignal.timeout(18_000),
          maxTokens: parsed.data.mode === "optimize" ? 2200 : 1600,
          model,
          prompt:
            parsed.data.mode === "optimize"
              ? `原始视频提示词：\n${structuredPrompt}\n\n请按照系统规则输出优化后的提示词正文。`
              : `待翻译的视频提示词：\n${structuredPrompt}\n\n请按照系统规则只输出译文。`,
          system: [definition.prompt, structureInstructions].filter(Boolean).join("\n\n"),
          temperature: getCompatibleTextGenerationTemperature(
            candidate,
            parsed.data.mode === "optimize" ? 0.65 : 0.25,
          ),
        } as any);
        const generatedText = normalizeGeneratedText(text);
        if (!generatedText) continue;
        const result = resolveVideoPromptStructure(
          generatedText,
          parsed.data.references,
          parsed.data.marks,
          { flatten: parsed.data.mode === "optimize" },
        );
        if (!result.text) continue;

        return NextResponse.json({
          success: true,
          data: {
            mentionOffsets: result.mentionOffsets,
            model: candidate.model.displayName,
            preservedMarkIds: result.preservedMarkIds,
            text: result.text,
          },
        });
      } catch (error) {
        logger.warn("prompt model candidate failed", {
          error,
          modelId: candidate.model.id,
        });
      }
    }

    return NextResponse.json(
      { success: false, error: "提示词处理模型暂时不可用，请稍后重试" },
      { status: 502 },
    );
  } catch (error) {
    logger.error("prompt tool failed", { error });
    const message = error instanceof Error ? error.message : "提示词处理失败";
    return NextResponse.json({ success: false, error: message }, { status: 500 });
  }
}