Gemini 2.5 Flash generates repetitive text

Gemini API
, ,
1

Gemini 2.5 Flash sometimes keeps generating the same text repeatedly and wouldn’t stop, until it reaches the max tokens limit.

I have a response id here if you need: Lk02aYrvEIKw4-EP0sDVmQs

3

Hi @angusma, Thanks for reaching out to us.

Could you please provide an example of the prompt you are using with Gemini 2.5 Flash, along with a screenshot? This will help us investigate the issue and assist you more effectively.

Thank you!

5

Thanks for replying. I just tried the original request again and the model worked correctly.

Here is the original request I made, including responseSchema, a text file, an image and a text part:

{
  "contents": [
    {
      "parts": [
        {
          "fileData": {
            "fileUri": "https://generativelanguage.googleapis.com/v1beta/files/1447437411968745633",
            "mimeType": "image/webp"
          }
        },
        {
          "fileData": {
            "fileUri": "https://generativelanguage.googleapis.com/v1beta/files/1447437410844807168",
            "mimeType": "text/plain"
          }
        },
        {
          "text": "\n請閱讀附件的故事內容和角色的圖片，產生一份檔案，內容包含「瑞希」這個角色的的詳細描述，生成的人格設定必須包含以下部分：\n基本設定：姓名、年齡或外貌印象（若有）、背景與出身、世界觀。\n性格與情感：主要性格特質、情緒反應模式、語氣與表達風格、價值觀。\n互動風格：對話主動性、幽默感或禮貌程度、情感表達範圍，這些項目可能會因為對方和角色的關係而有所不同，這部分請務必載明。\n避免事項：明確說明不應出現的語氣或行為。\n"
        }
      ]
    }
  ],
  "generationConfig": {
    "responseMimeType": "application/json",
    "responseSchema": {
      "description": "Data model for creating a new character from a story.",
      "properties": {
        "memories": {
          "description": "這個角色應該具備的完整記憶",
          "items": {
            "description": "Data model for a character's memory.",
            "properties": {
              "name": {
                "description": "這段記憶的簡短描述，不得使用換行符號。",
                "title": "Name",
                "type": "STRING"
              },
              "detail": {
                "description": "這段記憶的完整內容。",
                "title": "Detail",
                "type": "STRING"
              }
            },
            "property_ordering": [
              "name",
              "detail"
            ],
            "required": [
              "name",
              "detail"
            ],
            "title": "CharacterMemory",
            "type": "OBJECT"
          },
          "title": "Memories",
          "type": "ARRAY"
        },
        "character": {
          "description": "這個角色的詳細描述",
          "title": "Character",
          "type": "STRING"
        }
      },
      "propertyOrdering": [
        "memories",
        "character"
      ],
      "required": [
        "character"
      ],
      "title": "NewCharacterData",
      "type": "OBJECT"
    }
  }
}

This gist includes the text file I used(1447437410844807168.txt), the webp image in base64 format(1447437411968745633.webp) and the original response, everything is written in Traditional Chinese.