TikTok Mass Posting - Video Uniqueness, Metadata, TikTok Duplicate Detection Algorithm

hellowtf said:

Any updates,working with throung capcut or still stuck ?

Click to expand...

j3ffb3z0s said:

I was getting around 100 views on repurposed content and 200-500 on real content filmed in app with camera. This means the account is not shadowbanned, but I think my trustscore might be lower because I was posting from iPhone 7. On Monday I'm getting an iPhone 12 to test again.

Click to expand...

hellowtf said:

I just test mysefl with a viral video i reuploaded,deleted all metadata custom tags and leave the standard one and after i exported throung capcut to change metadata with capcut and i uploaded from phone galery Android s20+,seems in 9h i got 29 likes 2 add to fav and 700 wiews,seems it's not bad.

Click to expand...

hellowtf said:

I just test mysefl with a viral video i reuploaded,deleted all metadata custom tags and leave the standard one and after i exported throung capcut to change metadata with capcut and i uploaded from phone galery Android s20+,seems in 9h i got 29 likes 2 add to fav and 700 wiews,seems it's not bad.

Click to expand...

medex said:

How did you delete metadata bro

Click to expand...

craick said:

Tested it, works good!

Click to expand...

TheBusinessAlchemist said:

Making a video "unique" in the eyes of the TikTok algorithm is a reccuring topic here.

Many of us want to take a video, play with it's metadata and edit it with automation to have tens of duplicates ready to be posted, all unique in the eyes of TikTok.

In order to effectively bypass the TikTok Detection Algorithm, we need to understand and speculate about how it works and come up with potential fixes.

Let's use this thread to share all the tools and techniques we use to make our videos unique and the metadata we use to give more "trust" to the videos thus more reach.

Tools I use :

• ffmpeg : a tool to convert, apply filters, speed up, change metadata and modify in general all types of media files.
• python : a programming language to automate all my ffmpeg work in bulk. I use the "ffmpeg-python" library to make it easy to bulk edit thousands of videos with ffmpeg using python.
• exiftool : a command line tool to check metadata of all my videos. Usage : exiftool video.mp4 for example. Works on Unix based, not sure if available on Windows.

Potential TikTok Detection Methods :
I'm not an expert on the topic, it's all speculation, so if you have relevant info please share it below

• Hashing : the algorithm probably hashes every video (list of pixels) into a unique hash. If even one pixel of the video changes, the hash is completely different.
• Checking neighbor pixels : the algorithm probably checks the neighbor pixels of each video to see if, for example, the video has been slightly moved, cropped, or zoomed.
• Metadata : the algorithm probably checks if the metadata of the video is from a trusted source and if it unique or used thousands of times.
• File name : the algorithm maybe checks file name, comparing it to the typical file name used when editing on TikTok or CapCut to evaluate the trust of the video.

Potential Workarounds :

• Zooming in on the video : this method generates a zoomed duplicate of the original video, completely changing the hash. The pixels aren't the same anymore.
• Flipping the video : this method, while generating a unique hash, also completely bypasses a neighbor pixels algorithm.
• Speeding up the video : completely changes the hash. Not sure if very effective though.
• Removing metadata : this method at least makes the video "unique" to TikTok. Note: TikTok could consider this as a spam as very few original videos have no metadata.
• Adding trusted metadata : adding metadata generated by TikTok's editor or CapCut or some trusted metadata in general may help giving the video a boost of trust, TikTok thinking the video was manually edited on their platforms.
• Changing file name : changing file name to a typical file name generated by TikTok or CapCut.

Copy And Paste Code To Try FFMpeg Filters :

Python:

# Install ffmpeg, then run pip install ffmpeg-python

import ffmpeg
import os

# Put the videos you want to process in the "videos" folder
DIR = "videos"


def flip_video(path):
    try:
        (
            ffmpeg.input(path)
            .filter("hflip")
            .output(
                # Multiple ways to change metadata here
                # 1. map_metadata=-1 - removes all metadata generated by FFMpeg
                # 2. metadata="title=METADATA MAN"" - adds the metadata you want to add.
                # To add multiple metadata params, just use the syntax below
                path.replace(
                    ".mp4",
                    "_flipped.mp4",
                    # map_metadata=-1,
                    # metadata="title=METADATA MAN",
                    {
                        "metadata:g:0": "title=My Title",
                        "metadata:g:1": "artist=Me",
                        "metadata:g:2": "album=X",
                        "metadata:g:3": "year=2019",
                    },
                )
            )
            .run()
        )
        return True
    except ffmpeg.Error as e:
        print(e.stderr)
        return False


def main():
    files = os.listdir(DIR)
    for file in files:
        path = os.path.join(DIR, file)
        flip_video(path)


if __name__ == "__main__":
    main()

Improving Our Methods:
If you have any suggestion to help the many of us here to destroy the TikTok Detection Algorithm, post below and let's discuss other ways to get over the TikTok algorithm.

That's all, if you have any question (related or not to the code provided), don't hesitate to ask!

Click to expand...

Arsharca said:

Well, I don't know why it works for you.

Click to expand...

craick said:

Tested it, works good!

Click to expand...

Vlad2012 said:

Please test this "Video Cropping and Centering Script" and let me know

Code:

import cv2

input_video = "file name.mp4"  # Replace with the name of your video file
output_prefix = "output"
segment_duration = 300  # 5 minutes in seconds

cap = cv2.VideoCapture(input_video)
frame_width = int(cap.get(3))
frame_height = int(cap.get(4))
aspect_ratio = 9 / 16

fourcc = cv2.VideoWriter_fourcc(*'mp4v')
output_counter = 1
output_video = None

print("Starting video processing...")

while True:
    ret, frame = cap.read()
    if not ret:
        break

    # Calculate new dimensions for cropping while keeping the content centered
    if frame_width / frame_height > aspect_ratio:
        new_width = int(frame_height * aspect_ratio)
        left = (frame_width - new_width) // 2
        right = left + new_width
        new_frame = frame[:, left:right]
    else:
        new_height = int(frame_width / aspect_ratio)
        top = (frame_height - new_height) // 2
        bottom = top + new_height
        new_frame = frame[top:bottom, :]

    if output_video is None:
        output_video = cv2.VideoWriter(f"{output_prefix}{output_counter:03d}.mp4", fourcc, 30, (new_frame.shape[1], new_frame.shape[0]))

    output_video.write(new_frame)

    if cap.get(0) % (segment_duration * 1000) == 0:
        output_video.release()
        output_counter += 1
        output_video = None
        print(f"Segment {output_counter} generated...")

cap.release()
if output_video:
    output_video.release()

print("Video processing complete.")

1. Importing OpenCV (cv2) library:
   • The script starts by importing the OpenCV library, which is a popular library for computer vision and video processing tasks.
2. Setting input video and parameters:
   • The input_video variable is set to the filename of the video you want to process. You should replace "file name.mp4" with your actual video file.
   • The output_prefix is a prefix used for naming the output video segments.
   • The segment_duration is set to 300, which represents a segment duration of 300 seconds or 5 minutes.
3. Opening the input video:
   • The cv2.VideoCapture function is used to open the input video for processing.
   • The script retrieves the frame width and height from the video, which are used for cropping.
   • It also defines an aspect_ratio of 9/16, which is the desired aspect ratio for the output frames.
4. Configuring video output settings:
   • The script uses the cv2.VideoWriter_fourcc function to specify the video codec as 'mp4v' (MPEG-4 Video).
   • output_counter is initialized to 1, and output_video is set to None.
5. Starting video processing:
   • A message is printed to indicate that video processing is beginning.
6. Looping through video frames:
   • The script enters a while loop to process each frame of the input video.
   • It checks if the end of the video has been reached using the ret variable.
   • The script calculates new dimensions for cropping while keeping the content centered, ensuring the output frames have the desired aspect ratio.
7. Initializing and writing output video segments:
   • If output_video is None, it initializes a new video segment using cv2.VideoWriter.
   • The output frame is written to the current output segment.
   • If the current frame time is a multiple of the specified segment_duration, the current output segment is released, and a new one is started. The output_counter is incremented, and a message is printed to indicate a new segment is generated.
8. Releasing video resources:
   • After processing all video frames, the input video is released using cap.release().
   • If there is an open output video segment, it is also released using output_video.release().
9. Completing video processing:
   • A message is printed to indicate that video processing is complete.

The script essentially reads an input video, crops and resizes the frames to the desired aspect ratio, and creates multiple output video segments of a specified duration.

Click to expand...

j3ffb3z0s said:

I was getting around 100 views on repurposed content and 200-500 on real content filmed in app with camera. This means the account is not shadowbanned, but I think my trustscore might be lower because I was posting from iPhone 7. On Monday I'm getting an iPhone 12 to test again.

Click to expand...

j3ffb3z0s said:

I just made a script and posted the first video. Edits are quite minor but there's a lot of them. I'm hoping this passes the algorithm detection. Will update you guys in the morning. Here's a summary of the script (Generated with GPT):

Purpose:This script aims to process videos from a given directory (DIR), applying a series of visual and audio modifications. After processing, the videos are saved with metadata changes that make them appear as if they were recorded on a random iPhone model.

Imports:

• uuid: Used to generate a unique identifier.
• time: To get the current UNIX timestamp.
• os: For file and directory operations.
• ffmpeg: For changing video metadata.
• random: To make random choices and generate random numbers.
• numpy: Used for numerical operations on video frames.
• datetime: For time-related operations and formatting.
• moviepy.editor: To edit video files.
• moviepy.video.fx.all: For various video effects.
• moviepy.video.fx: For specific video effects like color modification.
• moviepy.audio.fx: For audio effects.

Functionalities:

1. Filename Generation: The generate_iphone_filename function creates a filename mimicking the structure often found with iPhone recordings.
2. iPhone Model Randomization: The random_iphone_model function returns a random iPhone model from a predefined list.
3. Lens Details: The lens_details_for_model function maps an iPhone model to its respective camera lens details.
4. Random US Location: The random_us_location function returns a random latitude and longitude within the continental US.
5. Portrait Orientation: The get_portrait_orientation function returns 90 degrees, indicating a portrait orientation.
6. Random Past Time: The random_past_time function gives a random timestamp from the recent past (up to 6 hours).

Video Modifications:

1. Mirror Effect: The video is mirrored horizontally.
2. Brightness Modification: Adjusts the brightness of the video.
3. Gamma Correction: Alters the video's gamma levels.
4. Random Pixel Modification: Randomly alters pixels in the video to a neutral gray color.
5. Zooming and Cropping: The video is zoomed in by a random factor and then cropped to its original dimensions.
6. Saturation Change: The video's saturation is adjusted.

Audio Modifications:The apply_random_audio_effect function divides the video's audio into segments. Each segment's volume is then randomly adjusted either up or down.

Metadata and Saving:

1. A new filename resembling iPhone naming conventions is generated.
2. The modified video is initially saved as an intermediate file.
3. The change_metadata function uses FFmpeg to change the metadata of this intermediate file. The metadata changes make it seem as if the video was recorded on a random iPhone model.
4. The video with the altered metadata is saved, and the intermediate video is deleted.

Main Execution:In the main function, the script iterates through video files in the specified directory (DIR), applies all the aforementioned modifications, changes their metadata, and then saves them in the converted_videos folder.

The aim of the entire process is to create videos that visually appear modified and have metadata resembling videos recorded on an iPhone.

Click to expand...

hunter2s said:

These are just basic video editing operations. TikTok's scrutiny of originality goes much further; currently, TikTok extracts one frame from every three to check for at least a 70% duplication before making a determination. This is also why many reposters get zero views. You need to learn deep deduplication, not just simple edits. The free software CR already achieves this and has more features, but it still cannot circumvent the issue of originality. Don't waste your time on this; you should consider exploring other directions

Click to expand...