| name | mimo-video-understanding |
| description | Analyze and understand video content using Xiaomi MiMo model — scene description, video summarization, and action detection with timestamps. Use this skill whenever the user shares or references a video file (.mp4, .mov, .avi, .wmv), a Base64-encoded video string, or a data URI (data:video/...;base64,...), even casually. Also use for: describing video content, summarizing tutorials or lectures, identifying actions in footage, extracting key moments, and analyzing screen recordings.
|
| license | MIT |
| metadata | {"version":"2.2","category":"ai-multimodal","sources":["https://platform.xiaomimimo.com/docs/zh-CN/usage-guide/multimodal-understanding/video-understanding"]} |
MiMo Video Understanding
Prerequisites
Requires mcp__mimo-multimodal__understand_video tool. If not available, read references/setup.md and help the user configure the MCP server — you will need to ask for their API plan type (token plan vs standard API) and credentials.
Supported Formats
MP4, MOV, AVI, WMV
- URL input: max 300MB per file
- Base64 input: max 50MB per encoded string
Input types: local file path, public URL, data:video/...;base64,... data URI, or raw Base64 string. The MCP tool auto-detects the input type and converts as needed.
Parameters
| Parameter | Default | Range | Description |
|---|
fps | 2 | 0.1 – 10 | Frames per second. Higher = finer detail, more tokens |
media_resolution | default | default, max | Per-frame resolution. max for small objects/textures |
FPS guide: 0.1–0.5 for lectures, 1–2 general, 3–5 tutorials, 5–10 sports/actions.
Analysis Modes
| Mode | When to use | Prompt |
|---|
| describe | General video understanding | "Describe this video in detail, scene by scene. Include: setting, people/objects, actions, any text or graphics overlaid, and overall context." |
| summarize | Long videos, tutorials | "Watch this video and provide a concise summary. Include: main topic, key moments, any instructions or demonstrations, and the overall purpose." |
| action-detect | Activity identification | "Identify and describe all distinct actions or events in this video. For each, note approximately when it occurs and what happens." |
Workflow
- Detect video inputs: local file paths (
.mp4, .mov, .avi, .wmv), public URLs, data:video/... data URIs, or raw Base64 strings
- Select the analysis mode and choose
fps / media_resolution based on content
- Call
mcp__mimo-multimodal__understand_video with the file path/URL and prompt
- Present results in the format below
Output Formats
describe: Scene-by-scene narrative description
summarize:
## Video Summary
### Main Topic
- ...
### Key Moments
1. [timestamp] — [what happens]
### Purpose
- ...
action-detect:
## Actions Detected
1. **[Action]** — ~[timestamp]
[Description]
Notes
- Video format variants are numerous — not all files are guaranteed to be recognized
- Audio tokens:
duration_seconds × 6.25; visual tokens depend on fps/resolution
- Use
media_resolution: "max" when fine details matter (small text, textures)
Error Handling
- 401 Unauthorized: API key is invalid or expired — ask user to check their key
- 413 Payload Too Large: Video exceeds 300MB (URL) or 50MB (Base64)
- Connection timeout: Check if
MIMO_API_BASE URL is correct for the user's plan
- Unrecognized format: Try converting to standard MP4 (H.264 codec)
For direct API calls (without MCP), see references/api-examples.md.
