Thread by @cyb_detective

Tools for video📽️ analysis

Basic online tools to help you use information from videos more effectively in your #osint investigations:

- text extraction
- deepfake detection
- instant identification of key frames
- metadata view

and more.

(1/15)🧵

Often investigators have to deal with videos that are posted on various sites. Download them to your computer using these services:

YouTube:
en.y2mate.is
Instagram, Facebook, Twitter, Tiktok: inflact.com/downloader/
Vimeo and others:
savethevideo.com/downloader

(2/15)🧵

For slow and thoughtful viewing of YouTube videos frame by frame, these services are handy:

watchframebyframe.com
anilyzer.com

And any online video editor can be used for frame-by-frame analysis of the video file:

Veed, Canva, Clipchamp, Kapwing etc

(3/15)🧵

And this service will help you quickly determine the timecodes on which there is a change of scenery in the video and significantly save time watching it.

huggingface.co/spaces/fffilon…

(4/15)🧵

If you can't understand what the people in the video are saying, then extract the text from it with neural networks, and then translate it with any translator.

YouTube:
huggingface.co/spaces/jeffist… (support 50+ languages)
Uploaded file:
huggingface.co/spaces/kabita-…

(5/15)🧵

You can also try just extracting subtitles from the video:

subtitlevideo.com
downsub.com

(6/15)🧵

When analyzing long videos, it is not necessary to read all of the recognized text (subtitles). You can just automatically generate a summary to quickly see what the point is:

huggingface.co/spaces/Gladiat…

(7/15)🧵

And, of course, it's possible to use ChatGPT to create a summary.

And you can use the Eightify extension to analyze YouTube video subtitles with ChatGPT:

chrome.google.com/webstore/detai…

(8/15)🧵

One of the main problems in investigations is identifying deepfakes. There are already a couple of tools for this:

deepware.ai
huggingface.co/spaces/Ron0420…

But these services are extremely inaccurate!
And frame-by-frame viewing is still the best method.

(9/15)🧵

Most often there is nothing interesting in the metadata of video files and it is useless to view them.

But there are occasional exceptions and it might make sense to check them.

flexclip.com/tools/metadata…

(10/15)🧵

An important part of video research is identifying faces, places, and objects from individual frames using reverse image search.

This is where image processing services come in handy:

remove.bg
watermarkremover.io
letsenhance.io

(11/15)🧵

In general, all the universal recommendations for reverse image search are also relevant when working with frames from video. I have a separate thread about this topic.

(12/15)🧵

https://twitter.com/cyb_detective/status/1548735521320587270

There are also special tools and techniques for researching #YouTube videos. Read more about them here:

(13/15)🧵

https://twitter.com/cyb_detective/status/1438405361954865154

When you watch a video, be aware of every unnatural position of objects or shadows. Today's neural network capabilities allow you to do incredible things. Here's an example of how easy it is to change soldiers from one uniform to another:
(14/15)🧵

https://twitter.com/cyb_detective/status/1625832273458102272

Thank you so much for reading this thread to the end!

If you want to receive a selection of my latest threads and articles in your inbox every 2-3 weeks, please subscribe to my Substack (it's free).

cybdetective.substack.com

(15/15)🧵

@threader compile
@threadreaderapp unroll
@threadrip unroll
@PingThread unroll
@threadreaders unroll
@TurnipSocial save
@readwiseio save thread
@tresselapp save thread
@rattibha unroll
@getnaked_bot unroll