Analysis of DeepNude Image Generation: Trends, Popularity

Here are the key points from this research:

1. Rapid Evolution: The DeepNude image generation technology has seen a rapid evolution, with more than 72 new solutions coming up in the past 16 months.
2. Massive Output: The total output of DeepNude images stands at an estimated 67-68.8 million images, with a monthly generation of around 3.3 million.
3. Platform Variation: The solutions are primarily web-based, but also include desktop applications and bots on Discord and Telegram.
4. Censorship Resistance: Despite censorship efforts, many of the blocked channels manage to restore their services quickly.
5. Changing Demand: A positive shift in demand has been noticed towards less explicit image generation.
6. Future Research: There’s an unexplored area concerning DeepNude videos and face-swapping in explicit content, which necessitates further research.

These key points provide a concise summary of the research and its implications for both the DeepNude image generation sector and the broader AI field.

Abstract:

This research provides a comprehensive overview of the DeepNude image generation landscape over the past 16 months, investigating the popularity, evolution, and daily generation volumes of such solutions. The study covers a review of over 70 different solutions, their quality of processing, speed, platform, pricing, and functionality. It also addresses the rising demand for less explicit image generation, a positive trend in an otherwise controversial field.

Introduction:

The world of AI-based DeepFake solutions, with its origins tracing back to the infamous DeepNude application, has seen a significant evolution over the past years.

DeepNude, a deep generative software based on image-to-image translation algorithm, excelling in undressing photos of humans and producing realistic nude images. The function of DeepNude is simple: input an image and generate the naked version of the image with a single click, consequence is catastrophic: anyone could now find themselves a victim of revenge porn. (1)

Despite its swift closure due to societal backlash, numerous solutions of this kind have emerged, prompting a necessity for detailed observation and understanding of this space. This study aims to address a significant knowledge gap, answering questions about the popularity of these solutions, their daily and total generation volumes of DeepNude photos, and their distribution based on the platform.

Trends

The Google Trends data for the search term “deepnude” shows several interesting insights:

1. Consistent Interest: The interest in the term “deepnude” has remained consistent over time, with only slight variations. This suggests that the topic is of ongoing relevance and interest to users worldwide.
2. Slight Decrease in Interest: Although the interest remains relatively consistent, the moving average suggests a slight decrease in interest over time. This could be due to various factors such as changes in news, societal trends, or the emergence of new related topics.
3. No Major Spikes or Drops: There are no major spikes or sharp drops in the search interest for this term, which indicates that there have been no significant events or news stories causing a sudden surge or decline in interest.

The Google Trends data for the search term “deepnude” in the US shows the following insights:

1. Lower Interest than Worldwide: The overall interest in the term “deepnude” in the US is lower compared to the worldwide interest. This suggests that this term may be less relevant or less known in the US compared to other regions globally.
2. Fluctuating Interest: Unlike the worldwide data, the interest in the US shows more fluctuations. There are periods of increased interest followed by periods of decreased interest. However, these fluctuations are not very drastic.
3. Decreasing Trend: The moving average line suggests a slight downward trend in interest over time. This is similar to the worldwide trend and indicates that interest in this term is slowly declining.

In summary, while the search term “deepnude” maintains a consistent level of interest worldwide, it appears to be less popular in the US, with a slightly more fluctuating pattern of interest. The declining trend in interest over time is also observed in both worldwide and US data.

Methods:

To provide a holistic perspective, this research involved monitoring and reviewing over 70 different DeepNude solutions that emerged over the observed period. The evaluation metrics included processing quality, operation speed, developed platform, pricing, and functionality. Furthermore, the platforms were categorized into three main types: websites (the most common), desktop solutions (requiring basic programming knowledge and limited due to restricted databases for training neural networks), and bots on Discord or Telegram.

Results:

Based on publicly available data from May 2023, over 33 million images have been generated. We calculated that, on average, one application accounts for over 300,000 image generations per month, which extrapolates to approximately 3 million images generated per month by ten applications of this scale. Desktop solutions cover an additional 2-7% of this quantity, adding approximately 60,000-210,000 images to the monthly total. Thus, the total DeepNude images generated per month, based on this analysis, could range from 3.06 to 3.21 million.

As of today’s date in July 2023, with the monthly generation rates remaining relatively steady since May 2023, we can estimate a further 2-2.1 million images have been produced (considering 1.5 months since the May data).

Therefore, the total DeepNude images generated until now could be in the range of approximately 67,560,950 images.

Discussion:

Despite controversies, this technology continues to evolve, showing a positive trend with increasing demand for less explicit image generation. However, one aspect that this study has not covered extensively is the generation of DeepNude videos and face-swapping in explicit photos or videos. The development of neural networks has made these solutions more accessible, but they are not yet as popular as their photographic counterparts.

Moreover, there have been instances of the blocking of such Telegram channels, with approximately one-third being quickly restored and continuing their active operation.

Conclusion:

This research offers valuable insights into the evolution, current state, and potential future trends of the DeepNude image generation sector. With technology advancing and societal attitudes shifting, this area is likely to continue its transformation, necessitating continuous monitoring and understanding of its implications.

References:

1. Yeh, Chin-Yuan, et al. “Disrupting image-translation-based deepfake algorithms with adversarial attacks.” Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision Workshops. 2020. https://openaccess.thecvf.com/content_WACVW_2020/papers/w4/Yeh_Disrupting_Image-Translation-Based_DeepFake_Algorithms_with_Adversarial_Attacks_WACVW_2020_paper.pdf
2. Lalonde, D. (2021). Policy Options on Non-Consensual Deepnudes and Sexual Deepfakes. Learning Network Brief 39. London, Ontario: Learning Network, Centre for Research & Education on Violence Against Women & Children. ISBN: 978-1-988412-49-8. https://www.vawlearningnetwork.ca/our-work/briefs/briefpdfs/Learning-Network-Brief-39.pdf
3. Google Trends https://trends.google.com/trends/explore?date=2020-01-01%202023-07-07&q=deepnude&hl=en-US

Author: Dr. Oleksandr Pedchenko, Ph.D., 07 July 2023, Tarragona, Spain.

Licensing Note:

This research paper is licensed under a Creative Commons Attribution License (CC BY). This license permits anyone to copy, distribute, display, and perform the work, including for commercial purposes, as long as the original author is credited appropriately.

Please cite this paper as follows:

Pedchenko, O. (2023). An Analysis of DeepNude Image Generation: Trends, Popularity. Outsource IT Today. Retrieved from https://outsourceit.today/analysis-of-deepnude-image-generation/