Research Study

Real Video Corpus Research Study

We are building a research corpus of authentic camera footage to study the signal patterns that distinguish genuine video from AI-generated or manipulated media.

This project focuses on technical signal analysis of authentic video.

Project overview

This study investigates the forensic signal patterns present in authentic camera-captured video. The goal is to improve authenticity analysis of video and better understand how real footage behaves after compression, editing, sharing, and platform processing.

By analysing these patterns across many clips, devices, and scenarios, we can build more reliable forensic tools for assessing whether a piece of video is genuine, manipulated, or synthetically generated.

Modern synthetic media — including AI-generated video — makes this increasingly important. Understanding what authentic footage actually looks like at a technical level is the foundation of reliable detection.

Why real footage matters

Many video detection systems are trained on clean or limited datasets. Real-world footage is far messier. It is resized, recompressed, cropped, screen-recorded, and shared through social platforms and messaging apps.

To understand genuine video properly, we need real camera-captured clips from the wild — recorded on real phones, in real lighting, with real compression, from real contributors. That diversity is what makes a forensic corpus reliable.

What kind of clips help

Short original clips captured by a phone, action camera, DSLR, mirrorless camera, or other real recording device are all useful.

B-roll and outtakes

Test footage and lens checks

Behind-the-scenes clips

Everyday phone recordings

Action camera footage

DSLR / mirrorless clips

Drone footage

Indoor and outdoor scenes

Clips do not need to be cinematic or polished. Short clips are useful. Even 2–3 clips from a single device can help. The only requirement is that the footage was captured by a real camera or phone — not AI-generated.

How footage is used

Uploaded clips are used strictly for forensic research analysis. We analyse technical characteristics of the video signal, including:

  • Compression artifacts and encoding patterns
  • Motion consistency and temporal signatures
  • Signal degradation through platforms and re-encoding
  • Device-specific signal fingerprints
Important: Uploaded footage is not used to train generative AI models. Original footage is not redistributed without permission.

Contributor credit

Contributors may choose to provide a website, portfolio, YouTube channel, or other URL. Contributors who do so can be acknowledged in the final study.

Where appropriate, backlink credit can be included for contributors who wish to be identified. Attribution is entirely optional — you can contribute anonymously if you prefer.

Privacy and responsible use

Please avoid uploading sensitive, highly private, or confidential footage. Clips should be safe to store for research purposes.

  • Clips are stored securely
  • The project uses footage for forensic analysis and technical reporting only
  • If any examples are referenced publicly, that is permission-based
  • Contributors can contact us to discuss or withdraw their submissions

How it works

1

Read the study details

Understand the purpose, what clips help, and how footage is used.

2

Upload original camera clips

Submit short real-world footage from any phone, camera, or recording device.

3

We analyse the signal patterns

Clips are processed for compression artifacts, encoding patterns, and motion consistency.

4

Contributors may be credited

If you choose to be identified, you can receive credit and a backlink in the final study.

Frequently asked questions

Ready to contribute footage?

Upload short original camera clips to help build a real-world video authenticity dataset. Every clip improves the quality of forensic analysis research.

Upload Your Clips

Contributions are voluntary. You can remain anonymous or receive credit in the final study.