# Does Video-Use Provide Playback Analytics? A Code-Level Analysis

> Discover if browser-use video-use provides playback analytics. This article reveals video-use is an LLM-driven video-editing engine, not an analytics tool.

- Repository: [Browser Use/video-use](https://github.com/browser-use/video-use)
- Tags: deep-dive
- Published: 2026-07-10

---

**Video-use does not offer any analytics features for video playback; it is architected exclusively as an LLM-driven video-editing engine focused on automated transcript-based cutting and quality control.**

The `browser-use/video-use` repository is a specialized tool for automated video editing, not a playback analytics platform. If you are evaluating this library for **video-use analytics** capabilities such as view tracking, engagement heatmaps, or watch-time metrics, the source code reveals a fundamentally different purpose. The entire codebase is designed around a transcription-to-edit pipeline that leverages large language models for content decisions rather than measuring viewer behavior.

## Architecture: Transcription and Visual Composite Layers

The video-use system operates through two distinct architectural layers that handle media processing, neither of which interfaces with playback telemetry.

**Audio Transcript Layer**: Each source video undergoes a one-time transcription via ElevenLabs Scribe, producing a compact, word-level transcript stored in [`takes_packed.md`](https://github.com/browser-use/video-use/blob/main/takes_packed.md). This file serves as the primary data structure that the LLM consumes to make editing decisions regarding cut points and content removal.

**Visual Composite Layer**: When the LLM requires visual context for ambiguous pauses or cut-point validation, the system invokes `timeline_view` from [`helpers/timeline_view.py`](https://github.com/browser-use/video-use/blob/main/helpers/timeline_view.py). This utility renders a film-strip, waveform, and word-label PNG for the specific time range requested, not for analytics visualization but for editing verification.

## The Editing Pipeline vs. Playback Metrics

The documented workflow follows a strict sequence:

```

Transcribe → Pack → LLM Reason → EDL → Render → Self-Eval

```

During the **Self-Eval** phase, the rendered output is re-examined using `timeline_view` to detect visual jumps, audio pops, or hidden subtitle issues. This feedback loop targets **quality control** of the edit itself, not the gathering of playback statistics. According to the [`README.md`](https://github.com/browser-use/video-use/blob/main/README.md), the feature set explicitly includes:

- Filler-word removal
- Auto-color-grading
- 30 ms audio fades
- Subtitle burning
- Animation overlay generation
- Self-evaluation of cuts

There is **no component that records, aggregates, or reports playback statistics** such as view counts, watch time, heatmaps, or user engagement metrics.

## Code Examples: Editing Utilities, Not Analytics

The core utilities demonstrate functionality strictly limited to editing and visual inspection. The following examples from the source code illustrate the library's actual capabilities:

```python

# Example: Generate a timeline view PNG for a given time range

from helpers.timeline_view import timeline_view

# Create a visual composite for seconds 30-45 of the source video

timeline_view(
    source_dir="my_videos",
    start=30.0,
    end=45.0,
    out_path="output/timeline_30_45.png",
)

```

This function generates static diagnostic images for the LLM's review, not interactive analytics dashboards.

```bash

# Example: Run the full editing pipeline on a folder of raw takes

uv run video-use

```

This command executes the editing skill referenced in [`SKILL.md`](https://github.com/browser-use/video-use/blob/main/SKILL.md), processing transcripts through the LLM reasoning layer to produce an Edit Decision List (EDL) for final rendering.

## Key Files Confirming the Scope

The repository structure reinforces the absence of analytics functionality:

- **[`README.md`](https://github.com/browser-use/video-use/blob/main/README.md)**: Documents the high-level pipeline and features without mentioning tracking or metrics infrastructure
- **[`helpers/timeline_view.py`](https://github.com/browser-use/video-use/blob/main/helpers/timeline_view.py)**: Generates on-demand film-strip and waveform composites used exclusively in self-evaluation
- **[`helpers/render.py`](https://github.com/browser-use/video-use/blob/main/helpers/render.py)**: Handles final video rendering after the LLM produces the EDL, with no hooks for telemetry
- **[`SKILL.md`](https://github.com/browser-use/video-use/blob/main/SKILL.md)**: Contains full production rules and editing craft guidelines, containing no references to data collection or viewer analysis

## Summary

- **Video-use** is strictly an LLM-driven video-editing engine, not an analytics platform.
- The architecture processes **transcripts** ([`takes_packed.md`](https://github.com/browser-use/video-use/blob/main/takes_packed.md)) and **visual composites** (`timeline_view`) for editing decisions, not viewer data.
- The **Self-Eval** loop inspects output quality using [`helpers/timeline_view.py`](https://github.com/browser-use/video-use/blob/main/helpers/timeline_view.py), focusing on visual/audio artifacts rather than engagement metrics.
- No components exist for tracking **view counts**, **watch time**, **heatmaps**, or **user engagement**.
- Analytics capabilities must be implemented separately or integrated via external services.

## Frequently Asked Questions

### Can video-use track viewer engagement or watch time?

No. The codebase contains no mechanisms for tracking viewer behavior, session duration, or engagement metrics. The `src` directory and helper modules focus exclusively on pre-render editing tasks such as transcript parsing and EDL generation. Any requirements for **video playback analytics** would require a separate instrumentation layer.

### What is the `timeline_view` function used for?

The `timeline_view` function in [`helpers/timeline_view.py`](https://github.com/browser-use/video-use/blob/main/helpers/timeline_view.py) generates diagnostic PNG images combining film-strips, waveforms, and word-level labels for specific time ranges. The LLM uses these composites to verify cut points and detect visual inconsistencies during the editing process, not to analyze how viewers interact with the final video.

### Does video-use integrate with external analytics platforms?

No native integrations exist. The repository is designed as a standalone editing tool that consumes raw video and produces finalized output through the `uv run video-use` command. It does not expose APIs, webhooks, or export formats compatible with analytics services like Google Analytics, Mixpanel, or custom telemetry dashboards.

### How does the self-evaluation feature work?

The **Self-Eval** stage re-examines the rendered output using `timeline_view` to catch quality issues such as visual jumps, audio pops, or hidden subtitle errors. This occurs after the `Render` step in the pipeline and serves as a final quality gate before delivery, functioning entirely as an automated editing assistant rather than a performance metric system.