Uniqueizing video with ffmpeg | My experience in creating YouTube Doorways

thrall said:

Hey man,

Im trying to create one "cartoon videos" for kids for me monetize it later...

I tried to edit one video with FFMPEG but i still getting copyright on youtube

I've used your tips:





Any other tip to help ?

Click to expand...

M4DM4X said:

Thanks mate for all your advice.
But you can make it easier. No need to change all those parameters, just use ffmpeg to reduce the screen size of the video and then insert a video background (like little stars floating).
For audio, also insert another track on top of the original track (reduce the volume).
The encoded result will be unique.
For increased safety you can also use ffmpeg to slightly distort the image (with a light convex filter). Their bots will think your video is unique.
Until it's flagged by humans complaining.

Click to expand...

mikenilox said:

Does it still work?

Click to expand...

FiveForwardSlash said:

This post was prepared as an answer to a question in this thread.
https://www.blackhatworld.com/seo/how-to-make-reposted-videos-unique.1337722/
But I decided to create a new one for the opportunity to supplement or make a series of posts on this topic of my experience with YouTube.
Everything written is my subjective experience and if you do not agree, please speak in the comments.

The topic with video uniqueness is very large and there won't be a simple answer here. But I will try to explain how it works and what settings you should try.

Globally, video verification in social networks and video hosting can be divided into two parts.
- Artificial intelligence, machine learning and live moderators. Typically, these methods are aimed at identifying inappropriate content (porn, blood, etc.) and copyrighted content.
- Methods that usually analyze the technical parameters of the video (including the hash sum).

I have good experience with the second methods, I will talk more about them. The work on changing the video code can also be divided into two important areas.

1. The first is the settings for recording the very information in the video. We are talking about rendering, video bitrate, audio bitrate, fame rate and constant rate factor (-crf).

2. The second direction is the change in the content of the video is imperceptible to the average user. Mirroring, Brightness, Contrast, Intro additions, gluing random parts of the video, etc.

Depending on the complexity of the task (the number of videos you want to download and the original ones), the number of settings will increase. Here are the tips arranged in order of increasing complexity of the task and the service you want to cheat.

1. Everywhere start by re-rendering your file. The fact is that this action requires little time and resources of the computer, but at the same time it allows you to get a rather unique hash sum without changing the content of the video, since the pixels will be overwritten in an arbitrary order.

2. Set a random Bitrate value. The most powerful tool in unique video is the bitrate, so it determines how much information will be recorded per second of the video and the weight of the final file. In my experience, it was this parameter that allowed us to achieve significant results. The more you want to get copies, the more you need to set the Bitrate range (for example, from 2M to 8M) with a step of 1M for each new video. 2M, 3M, 4M, 5M, 6M, 7M, 8M ...
For those who did not know, this parameter also accepts fractional numbers. For example, you can set the Video Bitrate "-b: v 6.5M".

3. Adding Brightness and Contrast. In this step, we are strengthening the 1st and 2nd method as we start to modify the content of the video but without much effort and harm to the video itself. The Brightness parameter (-vf eq = brightness =) takes a value from -1 to 1. I recommend the range from -0.2 to 0.2, since with large deviations from 0, the effects noticeably spoil the picture. In my templates, I set the generation of a random value to the recommended range with a step of 0.02. For example -0.2, -0.18, -0.16, -0.14 ...

The Contrast parameter (-vf eq = contrast =) takes values from -1000 to 1000. I recommend a range from -1 to 1. And a step of 0.2. -1, -0.8, -0.6, -0.4 ...

In my experience with YouTube, steps of 0.02 and 0.2 are sufficient to not be visible to the human eye, but significantly increase the uniqueness of the video in the eyes of the machine.

4. Change video length and add Intro. Using YouTube as an example, it turned out to significantly increase the number of copies from one file if we add a randomly prepared Intro from 4 seconds to 7 at the beginning of the video. For the intro, it is enough to make a picture, logo or an introductory animation (I did it in after effects). Make 3-5 of these and randomly glue to your main video. As a bonus, this will change the length of the video from the original.

This is enough to solve 90% of the problem and the final line would look like this.

Code:

ffmpeg -i "C:\Users\User\Desktop\Intro.mp4" -acodec copy -vcodec copy -vbsf h264_mp4toannexb -f mpegts "C:\Users\User\Desktop\Intro.ts"
ffmpeg -i "C:\Users\User\Desktop\Video.mp4" -acodec copy -vcodec copy -vbsf h264_mp4toannexb -f mpegts "C:\Users\User\Desktop\Video.ts"
ffmpeg -i "concat:C:\Users\User\Desktop\Intro.ts|C:\Users\User\Desktop\Video.ts" -s 1280x720 -b:v 6.5M -vf eq = brightness = 0.02 -vf eq = contrast = 0.8 "C:\Users\User\Desktop\VideoNew.mp4"

5. If this is not enough for your tasks then the next step is to generate a random Frame rate (-r) and Constant Rate Factor (-crf).

The value of the Frame rate should be set depending on the dynamics of your video, the power of your computer and the desired video weight. I do not recommend setting less than 30, as even with a slow slide show it looks noticeably bad. The higher the value, the more frames you need to render, which affects the render speed and the weight of the final file.

The -crf parameter takes values from 0 to 51 and it is he who is responsible for the video quality. But we are interested in the range from 18 to 28. 18 is better quality but more weight and 28 is worse quality but less file weight and the default value is 23.

Code:

ffmpeg -i "C:\Users\User\Desktop\Intro.mp4" -acodec copy -vcodec copy -vbsf h264_mp4toannexb -f mpegts "C:\Users\User\Desktop\Intro.ts"
ffmpeg -i "C:\Users\User\Desktop\Video.mp4" -acodec copy -vcodec copy -vbsf h264_mp4toannexb -f mpegts "C:\Users\User\Desktop\Video.ts"
ffmpeg -i "concat:C:\Users\User\Desktop\Intro.ts|C:\Users\User\Desktop\Video.ts" -s 1280x720 -r 30 -crf 28 -b:v 6.5M -vf eq=brightness=0.02 -vf eq=contrast=0.8 "C:\Users\User\Desktop\VideoNew.mp4"

Click to expand...

WebGeeksly said:

Does this still work? Might be worth testing, I will test this asap.

Click to expand...

FiveForwardSlash said:

This post was prepared as an answer to a question in this thread.
https://www.blackhatworld.com/seo/how-to-make-reposted-videos-unique.1337722/#post-14457374/
But I decided to create a new one for the opportunity to supplement or make a series of posts on this topic of my experience with YouTube.
Everything written is my subjective experience and if you do not agree, please speak in the comments.

The topic with video uniqueness is very large and there won't be a simple answer here. But I will try to explain how it works and what settings you should try.

Globally, video verification in social networks and video hosting can be divided into two parts.
- Artificial intelligence, machine learning and live moderators. Typically, these methods are aimed at identifying inappropriate content (porn, blood, etc.) and copyrighted content.
- Methods that usually analyze the technical parameters of the video (including the hash sum).

I have good experience with the second methods, I will talk more about them. The work on changing the video code can also be divided into two important areas.

1. The first is the settings for recording the very information in the video. We are talking about rendering, video bitrate, audio bitrate, fame rate and constant rate factor (-crf).

2. The second direction is the change in the content of the video is imperceptible to the average user. Mirroring, Brightness, Contrast, Intro additions, gluing random parts of the video, etc.

Depending on the complexity of the task (the number of videos you want to download and the original ones), the number of settings will increase. Here are the tips arranged in order of increasing complexity of the task and the service you want to cheat.

1. Everywhere start by re-rendering your file. The fact is that this action requires little time and resources of the computer, but at the same time it allows you to get a rather unique hash sum without changing the content of the video, since the pixels will be overwritten in an arbitrary order.

2. Set a random Bitrate value. The most powerful tool in unique video is the bitrate, so it determines how much information will be recorded per second of the video and the weight of the final file. In my experience, it was this parameter that allowed us to achieve significant results. The more you want to get copies, the more you need to set the Bitrate range (for example, from 2M to 8M) with a step of 1M for each new video. 2M, 3M, 4M, 5M, 6M, 7M, 8M ...
For those who did not know, this parameter also accepts fractional numbers. For example, you can set the Video Bitrate "-b: v 6.5M".

3. Adding Brightness and Contrast. In this step, we are strengthening the 1st and 2nd method as we start to modify the content of the video but without much effort and harm to the video itself. The Brightness parameter (-vf eq = brightness =) takes a value from -1 to 1. I recommend the range from -0.2 to 0.2, since with large deviations from 0, the effects noticeably spoil the picture. In my templates, I set the generation of a random value to the recommended range with a step of 0.02. For example -0.2, -0.18, -0.16, -0.14 ...

The Contrast parameter (-vf eq = contrast =) takes values from -1000 to 1000. I recommend a range from -1 to 1. And a step of 0.2. -1, -0.8, -0.6, -0.4 ...

In my experience with YouTube, steps of 0.02 and 0.2 are sufficient to not be visible to the human eye, but significantly increase the uniqueness of the video in the eyes of the machine.

4. Change video length and add Intro. Using YouTube as an example, it turned out to significantly increase the number of copies from one file if we add a randomly prepared Intro from 4 seconds to 7 at the beginning of the video. For the intro, it is enough to make a picture, logo or an introductory animation (I did it in after effects). Make 3-5 of these and randomly glue to your main video. As a bonus, this will change the length of the video from the original.

This is enough to solve 90% of the problem and the final line would look like this.

Code:

ffmpeg -i "C:\Users\User\Desktop\Intro.mp4" -acodec copy -vcodec copy -vbsf h264_mp4toannexb -f mpegts "C:\Users\User\Desktop\Intro.ts"
ffmpeg -i "C:\Users\User\Desktop\Video.mp4" -acodec copy -vcodec copy -vbsf h264_mp4toannexb -f mpegts "C:\Users\User\Desktop\Video.ts"
ffmpeg -i "concat:C:\Users\User\Desktop\Intro.ts|C:\Users\User\Desktop\Video.ts" -s 1280x720 -b:v 6.5M -vf eq = brightness = 0.02 -vf eq = contrast = 0.8 "C:\Users\User\Desktop\VideoNew.mp4"

5. If this is not enough for your tasks then the next step is to generate a random Frame rate (-r) and Constant Rate Factor (-crf).

The value of the Frame rate should be set depending on the dynamics of your video, the power of your computer and the desired video weight. I do not recommend setting less than 30, as even with a slow slide show it looks noticeably bad. The higher the value, the more frames you need to render, which affects the render speed and the weight of the final file.

The -crf parameter takes values from 0 to 51 and it is he who is responsible for the video quality. But we are interested in the range from 18 to 28. 18 is better quality but more weight and 28 is worse quality but less file weight and the default value is 23.

Code:

ffmpeg -i "C:\Users\User\Desktop\Intro.mp4" -acodec copy -vcodec copy -vbsf h264_mp4toannexb -f mpegts "C:\Users\User\Desktop\Intro.ts"
ffmpeg -i "C:\Users\User\Desktop\Video.mp4" -acodec copy -vcodec copy -vbsf h264_mp4toannexb -f mpegts "C:\Users\User\Desktop\Video.ts"
ffmpeg -i "concat:C:\Users\User\Desktop\Intro.ts|C:\Users\User\Desktop\Video.ts" -s 1280x720 -r 30 -crf 28 -b:v 6.5M -vf eq=brightness=0.02 -vf eq=contrast=0.8 "C:\Users\User\Desktop\VideoNew.mp4"

Click to expand...

FiveForwardSlash said:

Replying to DM

Click to expand...

Donsanti92 said:

I always get claimed using audio, I still didn't figure out the right commands to do, can you please share with me your latest findings?

Click to expand...

AmberJones said:

how about for youtube, will this work?

Click to expand...

WebGeeksly said:

Does this still work? Might be worth testing, I will test this asap.

Click to expand...

Juralgush said:

Hi,
Thanks for value.
Will your method still work in 2023? With the arrival of AI and improved algorithms, is it enough to change the hash sum and metadata for resposting on TikTok, YouTube, Instagram, etc? ? Or do we need to look for other methods?

Click to expand...

FiveForwardSlash said:

Try uniqueizing the audio first and only then add it. Try working with an equalizer (e.g. online mp3cut.net/equalizer), trim the length, add human speech to the protected music, or add audio tracks on top of dialog from protected movies. Use more than one track on top of protected dialog. Uniqueization of the final video is only 50% of success. The other 50% is the variety, quantity, uniqueness and complexity of the components you will assemble your video from before uniqueization.



YouTube is listed in the thread title. Or did I misunderstand the question?



The main idea remains the same. Add changes until the machine stops seeing matches (duplicates).
- I noticed that some spammers use only mirror reflections, rotations by a few degrees and stickers, but the number of copies is dozens or a couple hundred from dozens of different clips.
- I see a lot of effects created by neural networks on YouTube.
- Do it consciously creating control groups with different changes.
- Don't write that the method doesn't work if you managed to make only 90 copies without deletions, but make 10 unique videos and get your 900 copies)



Answered above. The hash sum shows only the fact of the smallest change, difference, interference. AI are now good helpers in uniqueization. Haven't worked with Instagram and TikTok but tried to help some forum members with TikTok without uploading myself. It turned out badly but I'm sure there is a solution if you do enough tests.

Click to expand...

Exhibit said:

I'm very interested in what you suggested with TT and found no success with.

Click to expand...

733899 said:

I tried these commands with some copyrighted clips and it did bypass the automatic copyright claim. Some of the videos are still active but few others were taken down manually from the copyright owner.

Click to expand...