Home
Creators
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register Favorites Keys Review DMs Logout
Community
Telegram ThePornDude
Home Creators Posts Import Register Favorites Logout

AI / tech / video editing people - advice needed! (Patreon)

Published:
2023-08-14 12:41:07
Imported:
2026-06

Content

I'm thinking of making a couple of videos for the channel along the lines of "every barbed wire fence from every straight line mission" or "Every time GeoWizard swore on a straight line mission" - those sorts of things. The problem is, there is well over a hundred hours of footage to trawl through and it simply wouldn't be worth the time and effort. 

So my question is, is there any AI software available to the public that can trawl through all the footage and identify specific types of clips? the audio of a swear word, the image of a barbed wire fence, etc etc. Let me know in the comments if you know of something that can help with this specific type of thing (Not asking for help with editing)By the way, for those wondering, there would be like 2 or 3 seconds of each clip so the video would hopefully not end up being too long.$4 Tier people: music stream and "hunted" video both just around the corner :)

Comments

Ben

Please do a bear grylls impression compilation!!

Daniel Noguera

Hey Tom, can certainly help with identifying any swear words via audio but the image recognition part will be a little more complicated.

Rob Unck

Davinci Resolve has a transcription feature with which you can ask it to make a transcription of a file or a timeline. You can then search that transcription for a word or a phrase and it will highlight in the timeline exactly where that word or phrase exists. I use the pro version at work, so I'm unsure if it's also available in the free version. I think even Premiere has transcription capabilities now, but I haven't tried it out yet. Let me know if you have any more Qs about this.

Simon Moos

I agree with Mark. Here's a service I found while googling which allows you to do audio/video to transcript: https://trint.com/.

Nuno Salvaterra

In another youtube channel I love, Cracking the Cryptic, there is someone in the comments that runs an AI to detect certain keywords in the audio. The user is https://youtube.com/@inspiringsand123 and you can find an example comment in https://youtu.be/RLNlZluaecM (scroll down until you see a comment with loads of timestamps starting with "Rules: 03:49". Maybe if you contact the user inspiringsand123 he can help you

George Hopkins

Hey Tom. You could use Davinci Resolve (free version) to generate automatic subtitles of all of the clips, and then search through that (Control-F - 'fuck', 'poop' etc.) If you already use Davinci then you can edit these directly, and if you use Premiere/FCP etc. then you can manually cut each clip out and in Davinci's export page, select 'individual clips' to export each of the clips of you swearing for use in other editing software. Another way (without Davinci) is to put all of your clips on an edit timeline and just export the audio. Then upload that audio into transcription software (Rev.com is great but $), then do the same search (Control F - 'wee', 'shit' etc.) You'll get timestamps from Rev (or whichever software you use for this) and then it's just a case of going through the edit timeline and picking the clips you want. The slickest, most up-to-date method is to use something like trint.com which is text-to-edit software, where you do the search (Control F - 'knobgobbler', 'willies' etc.) and it will edit it for you from the search. Not used it before but it's the future of editing (more time spent working on the story/video structure, less time spent shouting at your computer) Hope that helps. Let me know if I can clear anything up! George

Lukas Innig

On the image recognition: if someone can provide me with the videos sliced into images (once every few seconds) and a few hundred annotations, I have access to a ML platform that can then train on this data and score the whole vid. Some manual effort for the initial slicing and labeling is required

Lukas Innig

With annotations I mean "is this frame a fence or not"

Alex M.

you could download the generated close captions from Youtube and search through it. this way you get the timecode and you can cut the video. i think that is the easiest way for the cursing and swearing part 🤣

Nadav Horesh

You could hire a person on fiverr/upwork to manually do this for dirt cheap…

Eden Brent-Jones

Others have already answered about detecting certain said phrases (though I'd also add that you don't necessarily need to resort to paid versions of software, there are some decent transcription models that can be run locally virtually for free). For detecting certain images like barbed wire fences that's going to be more difficult unless there are already computer vision models out there that have been trained to detect the things you're looking for (I had a cursory search for ones that could detect barbed wire and couldn't find anything). Dogs or people (probably farmers in your case) might be an easier one to detect because there are certainly models out there already that can detect both. As someone who works with AI I'd love to say that there's an easy answer to your request but I'm tempted to say that crowd-sourcing manual efforts might be easier for the vision stuff!

bgr

are youtube's auto-generated subtitles usable?

Calum Stevens

I'd say fiverr is the way to go.

Bjørn Ravnås

Use Midjourney to make a new photo🤷‍♂️

David Priestley

Hey Tom. The swearing bit is fairly simple as others have mentioned - you just want to run the audio through a transcription service, then you can pick out the naughty bits from the output. You could even autogenerate a clip for each one and merge them together with something like moviepy in Python, but the timings might be a bit crappy. The video bit is much trickier but not impossible, you'd want to train a computer vision ML model on a load of images and essentially tell it whether each one contains a barbed wire fence or not, and then run your videos through it. You'll get peaks of scores around the frames that look most promising, and then it'd be a manual job to pick them out, because the accuracy won't be 100%. It might not save any time the first time you use it, but then you'd have it ready if you want to do something similar again.

Simon F.

You might want to have a look at YTKS (YouTube Keyword Search). I've not used it myself but heard good things. Of course this only helps with things you can find in the subtitles (like if you said the word "fence" for example). There is also a way to automatically download all the matching clips. https://ytks.app/

Gus Owen

Hi Tom, I have actually done this exact thing on your videos before, the link is here https://www.youtube.com/watch?v=mVc6cq4oGxY&t=44s . I do also have a whole other video of every time you have said 'get in' sitting on my hard drive but I haven't published it. To do it I wrote a little python script which gathers all your video transcriptions and then searches for a key word, and then automatically clips that part of the video, with a little manual tweaking. I'm also the bloke who made timeguessr.com which you did a couple videos on and to gather photos for that I sometimes use an ai to detect things like old cars (could be used for fences etc). If you want any help with this send me an email or contact me through here :)

Arsen C

I think this is what you might need: https://youtu.be/6-1V-QJEl-M Go through the video amd put a marker using the shortcut once you see it. It's a manual approach however, but its the easiest way and some guy on fiverr might be willing to do it

Peter Krämer

There is a website which let's you search videos for words that the subtitles contains. www.ytks.app You can tell it to go through your channel or straight line mission Playlist and to look for a specific word. It will show you the videos where the word was said with the according timestamp.

2xAA

Adobe’s Project Blink would be what you’re looking for (auto classification of shots not based purely on dialog), unfortunately they’ve closed beta sign ups for now. Could look at this instead? https://valossa.com/contextual-video-metadata/

Kalciob

I don't know about the bushes part, but you can search through the transcripts of your videos pretty easily with filmot.com, including any swear words.

louis boylan

The swearing is easier, you can use OpenAI’s whisperAI model to turn the spoken word into a text transcript. From there you can split up the text into chunks, labelled which video and which time span they represent for example. Then use a Python package such as Spacy to search the documents for the swear words (or what they might have been incorrectly transcribed to) and get a read out of the video and the time span in which it occurs. The barbed wire fences might be more difficult in that I expect there i not a good model for easily recognising that that is just straight off the shelf and even if there is one that you can fine turn the amount of data you would need to process would make it very expensive I expect. One solution could be to use the same approach as the swear words and search for words you might have said when a fence crossing is taken place such as “please not a wobbler” ;). That would at least give you a starting point. Honestly I am sure a barbed wire fence recognition model could be built (Google’s new segment anything model is pretty crazy) for visual images but it might just be more effort and trial and error than it’s saving versus skipping through the footage manually.

Andrea Sala

Hey tom, i had once seen a compilation video of all the times you said “get in” and I remember there was an explanation below the video on how the guy did it. I don’t have time at the moment to look it up but I think you should have a go!

geowizard

Thank you for your help everyone! Most useful!!

Serendipity.gallery

I could train you a custom model that detects moments where it sees bushes haha, but that will require you to first label hundreds of frames, and at this point the job of identifying the bushes will be done :p but it's doable

Andrew Weir

I'm not sure you need 'ai' for this. In Adobe Premiere, it'll transribe the audio and you can just search the transcription.

Tribal's Travels

I would have offered to go through it all personally and just edit it myself. (I'd charge though lol).

Travis

This isn't comment isn't related to this post, but this YouTuber called Ed Pratt is attempting to follow a river from its source to the sea, without leaving it. Reminds me of a mission you would do. Pretty cool, check it out. Maybe you can do something like that some day 👀

Dawnbandit

There's Whisper AI, which is an AI transcription/translation software. It has time stamps and everything and you can actually directly upload the .srt to Youtube. Also, it is extremely accurate and fast compared to Adobe's transcription.

geowizard

I've been meaning to do that for some time now! someone was bound to beat me to it the way ive dawdled, but yeah i still might!!

David Bennett Piano

Many people are mentioning auto-transcriber softwares which is a great idea… however you don’t even need to do that as YouTube automatically generates subtitles for your videos and you can download those subtitles as text files.