modelscope-image-mcp 1.0.2 → 1.0.3

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "modelscope-image-mcp",
-  "version": "1.0.2",
-  "description": "ModelScope (魔搭) 文生图 / 图生图 MCP 服务：免费调用 Qwen-Image、Z-Image、FLUX.1 等模型",
+  "version": "1.0.3",
+  "description": "ModelScope (魔搭) + SenseNova (商汤日日新) 文生图 / 图生图 MCP 服务：免费调用 Qwen-Image、Z-Image、FLUX.1、SenseNova U1.5 等模型",
   "type": "module",
   "main": "server.js",
   "bin": {
@@ -17,6 +17,7 @@
     "mcp",
     "model-context-protocol",
     "modelscope",
+    "sensenova",
     "text-to-image",
     "image-to-image",
     "image-editing",

package/server.js

@@ -13,7 +13,11 @@
  *   MODELSCOPE_OUTPUT_DIR  图片保存目录，默认 ~/modelscope-images
  *   MODELSCOPE_TIMEOUT_MS  任务超时毫秒数，默认 300000
  *
+ *   SENSENOVA_API_KEY      （可选）商汤日日新 API Key，配置后才能调用 sensenova-* 模型
+ *   SENSENOVA_API_BASE     商汤接口地址，默认 https://token.sensenova.cn/v1
+ *
  * 说明：零依赖，需要 Node.js >= 18（内置 fetch）。
+ *   魔搭走异步任务（提交 + 轮询）；商汤走同步接口（直接返回 base64），调用方无感。
  */
 
 import fs from "node:fs";
@@ -30,11 +34,21 @@ const OUTPUT_DIR =
 const TASK_TIMEOUT_MS = Number(process.env.MODELSCOPE_TIMEOUT_MS || 300000);
 const POLL_INTERVAL_MS = 3000;
 
+const SENSENOVA_API_BASE = (
+  process.env.SENSENOVA_API_BASE || "https://token.sensenova.cn/v1"
+).replace(/\/+$/, "");
+const SENSENOVA_API_KEY = process.env.SENSENOVA_API_KEY || "";
+
 const TEXT_TO_IMAGE_MODELS = [
-  { id: "Qwen/Qwen-Image", name: "Qwen-Image", desc: "通义千问图像生成，中英文文字渲染强，综合最稳" },
-  { id: "Tongyi-MAI/Z-Image-Turbo", name: "Z-Image-Turbo", desc: "轻量快速，出图速度快" },
-  { id: "Tongyi-MAI/Z-Image", name: "Z-Image", desc: "Z-Image 标准版，画质更好" },
-  { id: "black-forest-labs/FLUX.1-Krea-dev", name: "FLUX.1-Krea-dev", desc: "FLUX 系列，质感细腻" },
+  { id: "Qwen/Qwen-Image", name: "Qwen-Image", desc: "魔搭｜通义千问图像生成，中英文文字渲染强，综合最稳" },
+  { id: "Tongyi-MAI/Z-Image-Turbo", name: "Z-Image-Turbo", desc: "魔搭｜轻量快速，出图速度快" },
+  { id: "Tongyi-MAI/Z-Image", name: "Z-Image", desc: "魔搭｜Z-Image 标准版，画质更好" },
+  { id: "black-forest-labs/FLUX.1-Krea-dev", name: "FLUX.1-Krea-dev", desc: "魔搭｜FLUX 系列，质感细腻，中文 prompt 遵循差，建议英文" },
+  {
+    id: "sensenova-u1.5-lite",
+    name: "SenseNova U1.5 Lite",
+    desc: "商汤｜中文 prompt 直出效果好，默认无水印；不支持 negative_prompt，需配置 SENSENOVA_API_KEY",
+  },
 ];
 
 const IMAGE_TO_IMAGE_MODELS = [
@@ -135,12 +149,20 @@ function modelSlug(model) {
     .replace(/^-+|-+$/g, "");
 }
 
+/** 把图片字节写到输出目录，返回绝对路径。tag 是模型标识，会拼进文件名便于事后辨认 */
+async function saveImageBuffer(buf, ext, dir, tag = "") {
+  await fs.promises.mkdir(dir, { recursive: true });
+  const prefix = tag ? `ms-${tag}` : "ms";
+  const file = path.join(dir, `${prefix}-${timestamp()}-${Math.floor(Math.random() * 1e4)}${ext}`);
+  await fs.promises.writeFile(file, buf);
+  return file;
+}
+
 async function downloadImage(url, dir, tag = "") {
   const res = await fetch(url);
   if (!res.ok) throw new Error(`下载图片失败：HTTP ${res.status} ${url}`);
 
   const buf = Buffer.from(await res.arrayBuffer());
-  await fs.promises.mkdir(dir, { recursive: true });
 
   let ext = ".png";
   try {
@@ -150,10 +172,78 @@ async function downloadImage(url, dir, tag = "") {
     /* 忽略，用默认后缀 */
   }
 
-  const prefix = tag ? `ms-${tag}` : "ms";
-  const file = path.join(dir, `${prefix}-${timestamp()}-${Math.floor(Math.random() * 1e4)}${ext}`);
-  await fs.promises.writeFile(file, buf);
-  return file;
+  return saveImageBuffer(buf, ext, dir, tag);
+}
+
+// --------------------------------------------------------------- 商汤调用
+
+/** sensenova-* 走商汤同步接口，其余走魔搭异步任务 */
+function isSenseNova(model) {
+  return /^sensenova-/i.test(String(model));
+}
+
+/** 商汤 size 约束：32 的倍数、512–4096、宽高比不超过 3:1 */
+function checkSenseNovaSize(size) {
+  if (!size) return undefined;
+  if (String(size).trim() === "auto") return "auto";
+
+  const m = /^(\d+)x(\d+)$/.exec(String(size).trim());
+  if (!m) throw new Error(`商汤 size 格式应为 WIDTHxHEIGHT（如 1024x1024）或 auto，当前：${size}`);
+
+  const w = Number(m[1]);
+  const h = Number(m[2]);
+  const bad = [];
+  if (w % 32 || h % 32) bad.push("宽高都必须是 32 的倍数（如 1280x720 需改成 1280x736）");
+  if (w < 512 || w > 4096 || h < 512 || h > 4096) bad.push("宽高需在 512–4096 之间");
+  if (Math.max(w, h) / Math.min(w, h) > 3) bad.push("宽高比不能超过 3:1");
+  if (bad.length) throw new Error(`商汤 size 不合法（${size}）：${bad.join("；")}`);
+
+  return `${w}x${h}`;
+}
+
+async function runSenseNova({ model, prompt, size, dir, tag }) {
+  if (!SENSENOVA_API_KEY) {
+    throw new Error(
+      "未配置 SENSENOVA_API_KEY 环境变量：请在商汤日日新控制台获取 API Key，并在 MCP 配置的 env 中设置"
+    );
+  }
+
+  // watermark: false 关掉右下角的「日日新 sensenova」水印
+  const payload = { model, prompt, n: 1, watermark: false };
+  const checkedSize = checkSenseNovaSize(size);
+  if (checkedSize) payload.size = checkedSize;
+
+  const res = await fetch(`${SENSENOVA_API_BASE}/images/generations`, {
+    method: "POST",
+    headers: {
+      Authorization: `Bearer ${SENSENOVA_API_KEY}`,
+      "Content-Type": "application/json",
+    },
+    body: JSON.stringify(payload),
+  });
+
+  const text = await res.text();
+  let data;
+  try {
+    data = JSON.parse(text);
+  } catch {
+    throw new Error(`商汤生成失败：HTTP ${res.status} ${text.slice(0, 300)}`);
+  }
+  if (!res.ok) {
+    throw new Error(`商汤生成失败：HTTP ${res.status} ${JSON.stringify(data).slice(0, 500)}`);
+  }
+
+  const items = Array.isArray(data.data) ? data.data : [];
+  const files = [];
+  for (const item of items) {
+    if (item?.b64_json) {
+      files.push(await saveImageBuffer(Buffer.from(item.b64_json, "base64"), ".png", dir, tag));
+    } else if (item?.url) {
+      files.push(await downloadImage(item.url, dir, tag));
+    }
+  }
+
+  return { files, urls: [], model };
 }
 
 /** 把入参转成接口能接受的图片引用：http(s) 直接透传，本地文件转 data URI */
@@ -175,6 +265,15 @@ async function toImageRef(input) {
 // ------------------------------------------------------------------ 生成流程
 
 async function runGeneration({ model, prompt, negativePrompt, size, seed, image, outputDir }) {
+  const dir = outputDir || OUTPUT_DIR;
+  const tag = modelSlug(model);
+
+  if (isSenseNova(model)) {
+    if (image) throw new Error("商汤模型暂不支持图生图，请改用魔搭的图片编辑模型");
+    if (negativePrompt) log("提示：商汤接口不支持 negative_prompt，已忽略");
+    return runSenseNova({ model, prompt, size, dir, tag });
+  }
+
   const payload = { model, prompt };
   if (negativePrompt) payload.negative_prompt = negativePrompt;
   if (size) payload.size = size;
@@ -187,8 +286,6 @@ async function runGeneration({ model, prompt, negativePrompt, size, seed, image,
   const raw = result.output_images ?? result.output_image ?? [];
   const urls = (Array.isArray(raw) ? raw : [raw]).filter(Boolean);
 
-  const dir = outputDir || OUTPUT_DIR;
-  const tag = modelSlug(model);
   const files = [];
   for (const url of urls) {
     files.push(await downloadImage(url, dir, tag));
@@ -199,14 +296,14 @@ async function runGeneration({ model, prompt, negativePrompt, size, seed, image,
 
 function renderResult({ taskId, files, urls, model }) {
   if (!files.length) {
-    return `模型：${model}\n任务 ${taskId} 已完成，但未返回图片。原始响应：${JSON.stringify(urls)}`;
+    return `模型：${model}\n任务 ${taskId ?? "-"} 已完成，但未返回图片。原始响应：${JSON.stringify(urls)}`;
   }
   const lines = [`模型：${model}`, "", `已生成 ${files.length} 张图片：`, ""];
   files.forEach((f, i) => {
     lines.push(`${i + 1}. 本地文件：${f}`);
     if (urls[i]) lines.push(`   原始链接：${urls[i]}`);
   });
-  lines.push("", `任务 ID：${taskId}`);
+  if (taskId) lines.push("", `任务 ID：${taskId}`);
   return lines.join("\n");
 }
 
@@ -271,7 +368,7 @@ const TOOLS = [
   {
     name: "text_to_image",
     description:
-      "用魔搭（ModelScope）的文生图模型，按文字描述生成图片，自动保存到本地并返回文件路径。",
+      "文生图：用魔搭（ModelScope）或商汤日日新（SenseNova）的模型，按文字描述生成图片，自动保存到本地并返回文件路径。sensenova-* 模型走商汤同步接口（默认无水印）。",
     inputSchema: {
       type: "object",
       properties: {
@@ -280,8 +377,12 @@ const TOOLS = [
           type: "string",
           description: `模型 ID，默认 ${DEFAULT_TEXT_TO_IMAGE_MODEL}。可选：${TEXT_TO_IMAGE_MODELS.map((m) => m.id).join("、")}`,
         },
-        negative_prompt: { type: "string", description: "反向提示词，描述不希望出现的内容" },
-        size: { type: "string", description: "图片尺寸，如 1024x1024、1280x720（部分模型支持）" },
+        negative_prompt: { type: "string", description: "反向提示词，描述不希望出现的内容（商汤模型不支持，会被忽略）" },
+        size: {
+          type: "string",
+          description:
+            "图片尺寸，如 1024x1024、1280x720（部分模型支持）。商汤模型需为 32 的倍数且宽高比不超过 3:1，如 1280x736",
+        },
         seed: { type: "integer", description: "随机种子，便于复现同一张图" },
         output_dir: { type: "string", description: "图片保存目录，默认 ~/modelscope-images" },
       },
@@ -342,7 +443,7 @@ async function handleMessage(msg) {
       reply(id, {
         protocolVersion: params?.protocolVersion || "2024-11-05",
         capabilities: { tools: {} },
-        serverInfo: { name: "modelscope-image-mcp", version: "1.0.2" },
+        serverInfo: { name: "modelscope-image-mcp", version: "1.0.3" },
       });
       return;