Note: The gemini-pro-vision model (for text-and-image input) is not yet optimized for multi-turn conversations. Make sure to use gemini-pro and text-only input for chat use cases.

utils/common.js

@@ -1030,6 +1030,25 @@ export function getUserSpeaker (userSetting) {
   }
 }
 
+/**
+ * 获取或者下载文件，如果文件存在则直接返回不会重新下载
+ * @param destPath 相对路径，如received/abc.pdf
+ * @param url
+ * @param ignoreCertificateError 忽略证书错误
+ * @return {Promise<string>} 最终下载文件的存储位置
+ */
+export async function getOrDownloadFile (destPath, url, ignoreCertificateError = true) {
+  const _path = process.cwd()
+  let dest = path.join(_path, 'data', 'chatgpt', destPath)
+  const p = path.dirname(dest)
+  mkdirs(p)
+  if (fs.existsSync(dest)) {
+    return dest
+  } else {
+    return await downloadFile(url, destPath, false, ignoreCertificateError)
+  }
+}
+
 /**
  *
  * @param url 要下载的文件链接

utils/dalle.js

@@ -80,7 +80,7 @@ export async function imageVariation (imageUrl, n = 1, size = '512x512') {
   return response.data.data?.map(pic => pic.b64_json)
 }
 
-async function resizeAndCropImage (inputFilePath, outputFilePath, size = 512) {
+export async function resizeAndCropImage (inputFilePath, outputFilePath, size = 512) {
   // Determine the maximum dimension of the input image
   let sharp
   try {

utils/tools/SendPictureTool.js

@@ -19,6 +19,9 @@ export class SendPictureTool extends AbstractTool {
 
   func = async function (opt, e) {
     let { urlOfPicture, targetGroupIdOrQQNumber } = opt
+    if (typeof urlOfPicture === 'object') {
+      urlOfPicture = urlOfPicture.join(' ')
+    }
     const defaultTarget = e.isGroup ? e.group_id : e.sender.user_id
     const target = isNaN(targetGroupIdOrQQNumber) || !targetGroupIdOrQQNumber
       ? defaultTarget