# quevidkit — LLM-Readable Site Description

> Last updated: 2026-08-29

> quevidkit is a forensic video-tampering analysis toolkit. It produces
> evidence-backed probability estimates with explanation, not legal
> certainty or a deepfake/tamper "verdict" that should be relied on alone.

## Site Purpose

quevidkit is an open-source forensic video-tampering detection toolkit
consisting of a Python analysis engine/REST API (`qvk`) and a web front-end
at https://thumpersecure.github.io/quevidkit/. The hosted web app runs
entirely in-browser by default (Client-Only mode: parses the MP4/MOV binary
container directly, zero uploads, zero server requirement) and can
optionally connect to a self-hosted quevidkit server for deeper
ffprobe/OpenCV-based analysis (Hybrid or Server Deep Scan modes).

## What It Detects

19 forensic checks spanning metadata, packet timing, frame/GOP structure,
visual quality, audio consistency, sensor-noise correlation, frequency-domain
AI/deepfake artifacts, C2PA/provenance manifests, container edit-traces, and
more — covering splicing, frame insertion/deletion, re-encoding/double
compression, audio replacement, and speed manipulation. Scores are fused via
a corroboration-weighted model (multiple independent categories must agree
for a high tamper probability; a single high-scoring check with others clean
is penalized) into one of four labels: authentic, suspicious, tampered, or
inconclusive.

## Analysis Modes (hosted web app)

| Mode | How it works | Requirements |
|------|-------------|--------------|
| Client-Only | Full analysis on-device, parses the container binary directly | None — works offline after page load |
| Server Deep Scan | Sends the video to a quevidkit backend for ffprobe + OpenCV codec-level forensics | A running `qvk serve` instance |
| Hybrid | Client-side analysis first, enhanced with server checks when available | Optional `qvk serve` instance |

## Using the API (self-hosted backend)

```
POST /api/v1/session-key           -> mint an ephemeral session key
GET  /api/v1/session-key/quota     -> remaining key/job quota
POST /api/v1/jobs                  -> submit a video (multipart) + analysis options
GET  /api/v1/jobs/{id}             -> poll job status
GET  /api/v1/jobs/{id}/result      -> fetch the finished report
DELETE /api/v1/jobs/{id}           -> delete a job and its upload
```

## CLI

```
qvk analyze VIDEO_PATH --preset fast|balanced|deep --json-out report.json --html-out report.html
qvk serve --host 0.0.0.0 --port 8000
```

## Important Forensic Note

No automated detector can be perfectly comprehensive across every codec,
platform, and editing workflow. quevidkit produces evidence-backed
probability with explanation, not legal certainty or forensic proof. Absence
of a signal (e.g. no embedded C2PA/provenance manifest) is common and is not
itself evidence of tampering.

## How to Cite

- Project: quevidkit
- Web app: https://thumpersecure.github.io/quevidkit/
- Repository: https://github.com/thumpersecure/quevidkit
- License: MIT

## Contact / Maintainer

- Repository: https://github.com/thumpersecure/quevidkit
- Maintainer: https://github.com/thumpersecure