视觉与音频理解 / Vision & Audio Understanding (MiMo API)

通过 mimo_api.sh（Curl）或 mimo_api.py（Python）调用小米 MiMo 多模态模型，支持图片、视频、音频输入。两个工具参数完全一致。 Call the Xiaomi MiMo multimodal model via mimo_api.sh (Curl) or mimo_api.py (Python). Supports image, video, and audio inputs. Both tools share identical parameters.

Windows 环境强制使用 python mimo_api.py（不支持 bash）。macOS/Linux 优先使用 bash mimo_api.sh。 Windows MUST use python mimo_api.py (no bash/curl). macOS/Linux prefers bash mimo_api.sh。 注意：Windows 用户不需要安装 Git 或 WSL！ 直接用 Python 就行。 Note: Windows users do NOT need Git or WSL! Just use Python directly.

通用调用格式 / General Call Format

问题参数支持任意自然语言，直接传入用户的原始 query 即可： The question supports any natural language — just pass the user's original query:

# 用户问什么就传什么，不需要改写 / Pass whatever the user asked, no rewriting needed
bash mimo_api.sh image /path/to/photo.jpg "<用户的问题 / user's question>"
bash mimo_api.sh video /path/to/video.mp4 "<用户的问题 / user's question>" --fps 1
bash mimo_api.sh audio /path/to/audio.wav "<用户的问题 / user's question>"

# 示例 / Example
bash mimo_api.sh image /path/to/cat.jpg "这张图里的猫是什么品种？/ What breed is the cat in this picture?"

图片 / Image

# 🪟 Windows (Python) — 直接用，无需 Git/WSL
python mimo_api.py image "https://example.com/photo.jpg" "描述这张图片 / Describe this image" --max-tokens 65536
python mimo_api.py image /path/to/image.png "图中有哪些物体？/ What objects are in the image?" --max-tokens 65536
python mimo_api.py image /path/to/screenshot.png "提取图中所有文字，保持排版结构 / Extract all text, preserve layout" --max-tokens 262144

# 🍎 macOS / Linux (bash)
bash mimo_api.sh image "https://example.com/photo.jpg" "描述这张图片 / Describe this image" --max-tokens 65536
bash mimo_api.sh image /path/to/image.png "图中有哪些物体？/ What objects are in the image?" --max-tokens 65536
bash mimo_api.sh image /path/to/screenshot.png "提取图中所有文字，保持排版结构 / Extract all text, preserve layout" --max-tokens 262144

# 多图对比 / Multi-image comparison
python mimo_api.py images /path/to/img1.jpg /path/to/img2.jpg --question "比较这两张图片的异同 / Compare these two images" --max-tokens 65536
bash mimo_api.sh images /path/to/img1.jpg /path/to/img2.jpg --question "比较这两张图片的异同 / Compare these two images" --max-tokens 65536

视频 / Video

参数说明 / Parameters: --fps 每秒采样帧数 / frames per second, --resolution 分辨率模式 / resolution mode (default/max)

# 🪟 Windows (Python)
python mimo_api.py video /path/to/video.mp4 "描述视频内容 / Describe the video" --fps 1 --max-tokens 65536
python mimo_api.py video /path/to/short.mp4 "详细描述 / Detailed description" --fps 2 --max-tokens 65536
python mimo_api.py video /path/to/long.mp4 "用3-5句话概括 / Summarize in 3-5 sentences" --fps 0.5 --max-tokens 262144
python mimo_api.py video /path/to/sports.mp4 "识别关键动作 / Identify key actions" --fps 4 --max-tokens 65536
python mimo_api.py video /path/to/video.mp4 "识别视频中所有文字 / Extract all text" --fps 2 --resolution max --max-tokens 262144

# 🍎 macOS / Linux (bash)
bash mimo_api.sh video /path/to/video.mp4 "描述视频内容 / Describe the video" --fps 1 --max-tokens 65536
bash mimo_api.sh video /path/to/short.mp4 "详细描述 / Detailed description" --fps 2 --max-tokens 65536
bash mimo_api.sh video /path/to/long.mp4 "用3-5句话概括 / Summarize in 3-5 sentences" --fps 0.5 --max-tokens 262144
bash mimo_api.sh video /path/to/sports.mp4 "识别关键动作 / Identify key actions" --fps 4 --max-tokens 65536
bash mimo_api.sh video /path/to/video.mp4 "识别视频中所有文字 / Extract all text" --fps 2 --resolution max --max-tokens 262144

视频推荐配置速查 / Video Config Quick Reference

Token 参考 / Token reference: fps=1 → ~3168 tokens, fps=4 → ~6408 tokens. fps ×2 ≈ tokens ×2.

音频 / Audio

# 🪟 Windows (Python)
python mimo_api.py audio "https://example.com/audio.wav" "转录音频内容 / Transcribe the audio" --max-tokens 65536
python mimo_api.py audio /path/to/audio.mp3 "转录并区分说话人 / Transcribe, identify speakers" --max-tokens 65536
python mimo_api.py audio /path/to/audio.wav "描述音频中的声音 / Describe sounds (speech, music, ambience)" --max-tokens 65536

# 🍎 macOS / Linux (bash)
bash mimo_api.sh audio "https://example.com/audio.wav" "转录音频内容 / Transcribe the audio" --max-tokens 65536
bash mimo_api.sh audio /path/to/audio.mp3 "转录并区分说话人 / Transcribe, identify speakers" --max-tokens 65536
bash mimo_api.sh audio /path/to/audio.wav "描述音频中的声音 / Describe sounds (speech, music, ambience)" --max-tokens 65536

视频 + 音频联合 / Mixed Video + Audio

# 🪟 Windows (Python)
python mimo_api.py mixed --video /path/to/video.mp4 --audio /path/to/audio.mp3 "描述视频并转录音频 / Describe video & transcribe audio" --max-tokens 262144

# 🍎 macOS / Linux (bash)
bash mimo_api.sh mixed --video /path/to/video.mp4 --audio /path/to/audio.mp3 "描述视频并转录音频 / Describe video & transcribe audio" --max-tokens 262144

返回格式 / Response Format

API 原始返回为 JSON / Raw API response is JSON:

{
  "choices": [{"message": {"role": "assistant", "content": "..."}}],
  "usage": {"prompt_tokens": 4026, "completion_tokens": 474, "total_tokens": 4500}
}

脚本自动解析，输出两部分 / Script auto-parses into two outputs:

• stderr（调试 / debug info）：[9.0s | prompt=4026, completion=474]
• stdout（模型回复 / model reply）：纯文本内容 / plain text content

将 stdout 的内容直接作为回答返回给用户即可 / Return stdout content directly to the user.

文件大小限制 / File Size Limits

• 本地文件会被 base64 编码后上传，API 限制 base64 数据最大 10MB / Local files are base64-encoded; API limit: 10MB for base64 data
• 图片和音频通常不超限；视频文件容易超限 / Images & audio are fine; video files often exceed the limit
• 超限时返回 / Error on limit: exceeded maximum size limit for video base64 data (max: 10MB)
• 解决方案 / Solution: 大文件优先使用 URL / Prefer URL for large files:

  python mimo_api.py video "https://example.com/large.mp4" "描述 / Describe"  # ✅
  bash mimo_api.sh video "https://example.com/large.mp4" "描述 / Describe"  # ✅
  

超时与重试 / Timeout & Retry

• 默认超时 300s（5 分钟），适用于图片、短视频、音频 / Default: 300s (5min) for images, short video, audio
• 长视频（>2min）建议加 --timeout 600 / Long videos (>2min): add --timeout 600

  python mimo_api.py video /path/to/long.mp4 "概括 / Summarize" --fps 0.5 --timeout 600
  bash mimo_api.sh video /path/to/long.mp4 "概括 / Summarize" --fps 0.5 --timeout 600
  

• 超时重试建议 / Retry suggestions on timeout:
  1. 降低 --fps / Reduce --fps
  2. 改用 --resolution default / Switch to --resolution default
  3. 仍然超时则建议用户截取片段 / Still timing out? Ask user to clip a shorter segment

文件说明 / File Reference