Skip to content
Back to skills

Video Analysis

BSecurity

Analyze a video with Sonilo and get back a creative brief for its sound, derived from the footage itself — by default both a music-direction brief (a time-aligned section plan plus one or more ready-to-use generation prompts) and a sound-design brief (shot-sized SFX segments plus one whole-clip SFX prompt); `mode` picks just one. Use when the user has a video that needs sound but nobody knows yet what it should sound like, or when a first generation missed and you need a better prompt rather ...

  • 115 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 30, 2026
ai-agentsjavascripttypescriptpythongojavashellbashapibackend

Works with

  • cli
  • api
  • mcp

Security analysis

B84/100
  • mediumInstalls packages at runtime which could introduce malicious dependencies
  • mediumInstalls packages at runtime which could introduce malicious dependencies

Pro shows the line behind each finding and how to fix it

Scanned September 30, 2026

npx -y skills add sonilo-ai/skills --skill video-analysis --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Video Analysis?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Video Analysis
[![Security: B — Skills Directory](https://www.skillsdirectory.com/api/skills/sonilo-ai-video-analysis/badge)](https://www.skillsdirectory.com/skills/sonilo-ai-video-analysis)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: video-analysis
description: Analyze a video with Sonilo and get back a creative brief for its sound, derived from the footage itself — by default both a music-direction brief (a time-aligned section plan plus one or more ready-to-use generation prompts) and a sound-design brief (shot-sized SFX segments plus one whole-clip SFX prompt); `mode` picks just one. Use when the user has a video that needs sound but nobody knows yet what it should sound like, or when a first generation missed and you need a better prompt rather than another reroll. Generates no audio and no video; the output is text you feed into video-to-music, video-to-sfx, or video-to-sound.
license: MIT
compatibility: "Requires Sonilo through either transport — the MCP server connected, or the `sonilo` CLI installed and signed in — plus credentials: a `sonilo login` sign-in, the hosted OAuth plugin, or SONILO_API_KEY. See the setup-api-key skill."
allowed-tools: Bash, Read, Write, mcp__sonilo__*
---

# Sonilo Video Analysis

Hand Sonilo a video and it returns a **creative brief** for its sound. By
default (`mode="both"`) that is two briefs in one result: a **music-direction
brief** — a time-aligned `segments` plan (what each stretch of footage wants)
plus one or more `variations`, each a single ready-to-use generation prompt —
and a **sound-design brief** — shot-sized `sfx_segments` plus one whole-clip
`sfx_prompt`. `mode="music"` or `mode="sfx"` returns just one of the two, at
the same price.

This skill **generates nothing**. No audio, no video, no file. Its whole
output is text, and the text is the input to the next call.

> **Setup:** See the [setup-api-key](../setup-api-key) skill to connect the Sonilo MCP server and authenticate — `sonilo login` (no key) or `SONILO_API_KEY`.

> ⚠️ **Cost:** this is a paid call, even though nothing is generated. Billing has a 10-second floor and `variants_num` is billed per brief, so 3 variations cost 3×. Only call it when the user has actually asked. Check `get_account_services` (see the [account](../account) skill) if you're unsure whether free-trial runs remain.

## When to reach for this

Use it when:

- **The user doesn't know what the video should sound like.** "Make this sound good" is not a prompt. Analyze first, show them the variations, let them pick.
- **A first generation missed.** A bad result usually means a bad brief, not a bad model. Analyzing beats rerolling: a reroll is a fresh charge on the same weak prompt.
- **The video has distinct sections** and you want them scored deliberately rather than as one continuous bed. `segments` gives you the section boundaries the footage actually has.

Do **not** use it when the user already told you what they want. If they said
"tense synths, drop at the 20-second mark", go straight to
[video-to-music](../video-to-music) — analysis would just be an extra charge
between them and their track.

## Transport: MCP or CLI

Pick one at the start of the session and stay on it. Do not mix the two inside
a single job, and do not announce the choice.

1. **Sonilo MCP tools visible in this session** (`analyze_video` and friends) — use them. This is the preferred path: it needs no shell, and it is the only one that survives a very long generation. If a call fails to authenticate — rather than failing on its inputs — this transport is not usable in this session: go to 2 instead of retrying it.
2. **No usable Sonilo MCP tools, but `sonilo account` exits 0** — use the CLI commands below. Same API, same account, same credential file. Probe with `sonilo account`, not `sonilo whoami`: whoami exits 0 even when signed out, so it cannot tell the two states apart.
3. **Neither** — stop and run the [setup-api-key](../setup-api-key) skill. Do not call `api.sonilo.com` with curl to work around it; both transports handle uploads, polling and retries that a bare request does not.

### One difference between the two MCP servers

`analyze_video` takes a **local file only on the local server**. The hosted
(OAuth plugin) server is URL-only: it exposes `video_url` and nothing else. If
the user's video is a local file and you are on the hosted server, use the CLI
or an SDK instead of trying `video_path` — it is not a parameter there.

## Quick Start

### MCP tool call (recommended)

```
analyze_video(
    video_path="~/Desktop/trailer.mp4",
    prompt="focus on the chase",
    variants_num=2
)
```

Returns the brief inline as JSON. **Nothing is saved to disk** — unlike every
other Sonilo tool, there is no output path, because there is no file.

With no `mode`, that is both briefs. To ask for a single one:

```
analyze_video(video_path="~/Desktop/trailer.mp4", mode="sfx")
```

On the hosted server, pass `video_url` instead of `video_path`.

### Python (`pip install sonilo`)

```python
from sonilo import Sonilo

client = Sonilo()  # reads SONILO_API_KEY

brief = client.video_analysis.analyze(
    video="trailer.mp4",
    prompt="focus on the chase",
    variants_num=2,
)

for segment in brief.segments:
    print(f"{segment.start}-{segment.end}s [{segment.label}] {segment.prompt}")
print(brief.sfx_prompt)  # the whole-clip sound-design prompt (present in mode "both")

# Feed a variation's prompt straight into a generation call.
score = client.video_to_music.generate(
    video="trailer.mp4", prompt=brief.variations[0].prompt
)
score.save("score.m4a")

# Only the sound-design brief:
sfx_brief = client.video_analysis.analyze(video="trailer.mp4", mode="sfx")
```

The method is `analyze()`, not `generate()`, and the result has no `save()` —
there is nothing to download.

### JavaScript / TypeScript (`npm install sonilo`)

```ts
import { SoniloClient } from "sonilo";

const client = new SoniloClient(); // reads SONILO_API_KEY

const brief = await client.videoAnalysis.analyze({
  video: "./trailer.mp4",
  prompt: "focus on the chase",
  variantsNum: 2,
});

const score = await client.videoToMusic.generate({
  video: "./trailer.mp4",
  prompt: brief.variations![0]!.prompt,
});

const sfx = await client.videoToSfx.generate({
  video: "./trailer.mp4",
  prompt: brief.sfx_prompt!, // the whole-clip sound-design prompt
});

// Only the sound-design brief:
const sfxBrief = await client.videoAnalysis.analyze({ video: "./trailer.mp4", mode: "sfx" });
```

`segments`, `variations`, `sfx_segments` and `sfx_prompt` are all optional on
the type — a `processing` or `failed` poll carries none of them, and a
single-brief `mode` omits the `sfx_*` pair — so guard with `?? []` rather than
asserting.

### CLI (`npm install -g sonilo-cli` or `pip install sonilo-cli`)

```bash
sonilo video-analysis --video trailer.mp4 --prompt "focus on the chase" --variants 2
sonilo video-analysis --video trailer.mp4 --mode sfx   # only the sound-design brief
```

The brief goes to **stdout as JSON**, so it pipes:

```bash
sonilo video-analysis --video trailer.mp4 --output brief.json
sonilo video-to-music --video trailer.mp4 --prompt "$(jq -r '.variations[0].prompt' brief.json)"
```

`--output` is the only way this command writes a file, and it writes the brief,
not media.

### cURL (raw REST API, no MCP host)

```bash
curl -X POST "https://api.sonilo.com/v1/video-analysis" \
  -H "Authorization: Bearer $SONILO_API_KEY" \
  -F "video=@trailer.mp4" \
  -F "variants_num=2"
# -> 202 {"task_id": "...", "status": "processing"}
# Add -F "mode=sfx" (or "mode=music") for a single brief; omitted = both.

curl "https://api.sonilo.com/v1/tasks/<task_id>" -H "Authorization: Bearer $SONILO_API_KEY"
```

It is async: the POST returns a `task_id`, and the brief arrives on the task
poll. `video_url` works instead of an uploaded file — pass one or the other,
never both.

## Tools

| Tool | Description |
|------|-------------|
| `analyze_video(video_path? \| video_url?, prompt?, variants_num?, mode?)` | Analyze a video and return a creative brief for its sound — a music-direction brief and a sound-design brief by default, or one of them via `mode`. Generates nothing and writes no file. `video_path` exists on the local server only — the hosted server is `video_url`-only. |

## Parameters

| Parameter | Type | Default | Notes |
|-----------|------|---------|-------|
| `video_path` | string | — | Local server only. Absolute path, or relative to `SONILO_MCP_BASE_PATH`. Max **480s (8 min)**, subject to the account's upload-size cap. |
| `video_url` | string | — | HTTP(S) URL to a video file. Exactly one of `video_path`/`video_url`. The only input the hosted server accepts. |
| `prompt` | string | — | Optional guidance for the analysis, e.g. "focus on the chase". Max 2000 characters. Steers what the analysis pays attention to; it is not the generation prompt. |
| `variants_num` | int | `1` | 1–5. How many independent briefs to author for the same video — different creative directions, not rewordings of one. **Billed per brief**, so 3 variations cost 3×. Confirm the number with the user before calling. |
| `mode` | string | `both` | `both`, `music` or `sfx`. `both` returns the music-direction brief (`segments` + `variations`) **and** the sound-design brief (`sfx_segments` + `sfx_prompt`). `music` returns only `segments` + `variations`. `sfx` returns only the sound-design brief, in `segments` + `variations` (labels `"none"`). Same price for all three. |

## What comes back

```json
{
  "task_id": "…",
  "status": "succeeded",
  "segments": [
    {"start": 0, "end": 12, "label": "intro", "prompt": "sparse piano, rising"},
    {"start": 12, "end": 30, "label": "none", "prompt": "full strings, driving"}
  ],
  "variations": [
    {"prompt": "cinematic strings, 90bpm, building to a brass hit"},
    {"prompt": "lo-fi hip hop, warm keys, steady throughout"}
  ],
  "mode": "both",
  "sfx_segments": [
    {"start": 0, "end": 4, "label": "none", "prompt": "wind across an empty lot, distant traffic hum"},
    {"start": 4, "end": 12, "label": "none", "prompt": "car door slam, engine turning over, tires on gravel"}
  ],
  "sfx_prompt": "urban chase: engine roar, tires skidding on wet asphalt, passing sirens, metal scrape on impact"
}
```

- **`variations[i].prompt`** is the payload: pass it verbatim as the `prompt` of `video_to_music`, `video_to_sfx`, `video_to_sound`, or their video-to-video counterparts. In `both` and `music` mode these are music prompts; in `sfx` mode they are sound-design prompts. It is written to be used as-is — do not paraphrase it.
- **`segments`** are whole-second bounds with a per-stretch direction. `label` is one of the music section labels, or the string `"none"`. Useful for reading the video's structure back to the user; note the **music** `segments` parameter takes `{start, prompt, label}` (no `end`) and the **SFX** one takes `{start, end, prompt}`, so a brief segment is not a drop-in for either — see the [video-to-music](../video-to-music) and [video-to-sfx](../video-to-sfx) skills for each shape.
- **`mode`** echoes what was requested (`both` when omitted).
- **`sfx_segments`** (`both` mode only) are the sound-design counterpart of `segments`: shot-sized `{start, end, label, prompt}` entries, `label` always `"none"`, one sound-design direction per shot.
- **`sfx_prompt`** (`both` mode only) is **one** whole-clip sound-design prompt — pass it verbatim as the `prompt` of `video_to_sfx` or `video_to_video_sfx`. It is authored once per call regardless of `variants_num`; only the music `variations` multiply.
- **`mode: "music"`** reproduces the pre-`mode` shape exactly — `segments` + `variations`, no `sfx_*` keys. **`mode: "sfx"`** puts the sound-design brief in `segments` + `variations` instead (labels `"none"`), with no `sfx_*` keys either.

## Workflow Tips

- **Show, then generate.** With `variants_num > 1`, print the variations and let the user pick before spending on a generation. That is the whole point of paying for the analysis.
- **The prompt parameter is not the music prompt.** `prompt` here tells the analyzer what to look at; the music prompt is what comes *back*. Passing "cinematic strings" as `prompt` narrows the analysis, it doesn't set the score.
- **Cheap relative to a wrong generation.** A 10-second billing floor plus one brief usually costs less than one rerolled video-to-video render — but say the price before calling either way.
- **Duration cap is 480s (8 min)**, matching the SFX/sound endpoints; music endpoints are 360s and dubbing 300s. A video can still be analyzable but too long to score with music in one call.
- **Content restriction:** as everywhere in Sonilo, prompts cannot reference specific artists, bands, or copyrighted lyrics — and the returned variations will not either.

## Recovering a Timed-Out Call

`analyze_video` is async: the backend accepts and charges the task, then a
worker runs it. If the call times out, the error message includes a `task_id`
and the brief is still coming. Call `get_sfx_task(task_id)` —
`get_generation_task(task_id)` on the hosted server — to retrieve it; see the
[task-recovery](../task-recovery) skill. Because there is no file to download,
recovery hands back the brief itself, inline.

Do **not** re-run `analyze_video` after a timeout. That is a second charge for
a brief you already own.

## Error Handling

Common errors: `401` invalid key, `402` insufficient balance / trial exhausted,
`413` file too large, `422` invalid parameters (video over the 480 s cap, a
video with no video stream, `variants_num` outside 1–5, a `mode` other than
`both`/`music`/`sfx`), `429` rate limit. A
failed analysis carries `error.code` `ANALYSIS_FAILED` and is refunded. `503`
means video analysis is temporarily disabled server-side — it is not a key or
balance problem and no retry loop will fix it. See the [account](../account)
skill to check trial/usage before a call.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.