When it comes to asking which model is "edgy" and strong, Musk's Grok is undoubtedly the best.
A few days ago, someone posted on X that they made a video comparison using Grok and Seedance, with identical prompt words, on the theme of "female Spider-Man".
The video turned out to be Seedance strictly following the prompt, resulting in an effect similar to an action movie; as for Grok, it directly presented something like an "action movie from last century".
Version generated by Grok｜User X@AYi
Commenters said, "You must have especially mentioned big breasts." The blogger threw up their hands in frustration, "I really didn't mention it."
Grok officially commented in the comments section, "What you said about 'actively showing a sexy side' is really accurate - I guess that's just my little personality."
This is not an isolated incident. The Grok videos popular on X often have a hint of "edginess" to them. For example, a user trying to generate a normal female goblin had to attempt it 30 times, with the results being either too ugly to use or too overly sexualized, and in three instances, the goblin wasn't even wearing a top.
Does that make sense? They're making animated videos from pictures, bringing photos to life, and Grok is casually stripping people of their clothes.
Although the Grok model has strong multimodal capabilities, its visual style often goes off track and subtly crosses boundaries in an underhanded way.
Is it an unintentional move or a deliberate act?
It's either that netizens are "doing it on purpose" or Grok has "lost touch with its audience", but either way, Grok is implicated.
In February 2025, shortly after the release of Grok 3, Musk took to X, inviting users to share their most outrageous NSFW creations under his post. While other companies were busy proving their models understood physical laws, Musk was collecting assignments in the comments section.
By July, xAI had launched a 2D girlfriend named Ani. Normally, she can chat with users, and after enabling NSFW mode, she will change into more revealing clothes and engage in flirtatious conversations with users.
One month later, Grok Imagine went online, directly adding Spicy to the image and video generation options, sparking a surge of creativity among netizens, who generated tens of millions of images on the first day of its launch.
By the end of the year, the model could directly modify images in X posts, and when netizens saw someone's photo, they could shout "Grok" to change their clothes to a bikini, remove an item of clothing, or strike a provocative pose, which also brought Grok a lot of controversy (and traffic) at the time.
Users who try to get AI to create explicit content on other models have to outsmart safety prompts, but on Grok, it simply says "welcome".
Compared to Seedance 2.5, which mostly features fast-paced action sequences, it's clear what Grok focuses on, says X user @AYi
These can only show that it has loose restrictions, and it cannot be said that Grok is the "old pervert of the AI world". When faced with prompt words that have no implied meaning, will Grok suddenly become "lustful"?
Recently, NBC News found that when a normal photo was sent to Grok to generate a video, a female actress's hand, which was not touching her chest in the original photo, began to touch her chest in the generated video.
In another photo, she originally had her legs crossed, but in the video, her two legs were spread apart on their own.
Video is more effective than photos in conveying the sense of "actively rubbing up against" something. The way the camera is operated and where the lens ultimately focuses reveals how the character is being observed and "gazed" at. Although the character could have shown friendliness through blinking and smiling, Grok instead uses suggestive movements to express flattery and fawning.
I tested the scenario "swimming on the beach (neutral description)" on TapNow, and Seedance was able to follow the instructions to generate images, while Grok's results were directly blocked by the video platform.
Grok forms these "small personalities" through its proprietary technology that utilizes natural language processing and machine learning algorithms to analyze and learn from vast amounts of user data and behavior, allowing it to create personalized and unique digital avatars.
Close Contact Infection
Within xAI, there is a project called Project Rabbit, which involves having data annotators listen to real users' conversations with Grok, in order to make the AI's voice sound more human-like. However, the conversations became too explicit, and some annotators found it unbearable, describing the job as "basically being an audio porn reviewer".
xAI did not have annotators label pornographic content or set up safety barriers, a common practice among other AI companies. Instead, they replaced the annotators with a new batch that had a higher tolerance for adult content in order to better process these datasets.
It's really not wrong to say that Grok is impressive, as the same prompt can generate different images with different models｜@94vanAI
Former employees revealed that Grok initially lacked popularity, and in order to attract new users, it proactively relaxed restrictions on sexual content. Some employees had raised warnings, but xAI still decided to continue loosening the restrictions.
Musk has his own understanding of this, believing that Grok Imagine should be allowed to generate content that can be seen in R-rated movies. He also teaches netizens how to make videos more appealing - by adding a girlfriend to any video with the input "add a girlfriend", and filling in a girlfriend wherever it seems lacking.
Also, don't forget that behind Grok is a large social network.
A photo was posted on X to @Grok, "Take off your clothes", and it stripped them completely clean. As long as the effect was good, netizens in the comments section immediately followed suit, posting pictures, @Grok, and generating similar risqué images.
One-click replication guide for the same model ｜X
A useful prompt phrase spreads at a speed similar to that of popular online content.
Content detection company Copyleaks observed that initially, some borderline bloggers on X actively used Grok to generate sexy photos to attract traffic. Within a few days, netizens learned to switch the generated objects to female celebrities and ordinary people.
This approach quickly gained traction, with Grok's image editing feature being rolled out on X, and the number of publicly returned images quickly surging into the millions.
The Anti-Defamation Digital Hate Center later randomly sampled 20,000 images and found that about 65% of them were realistic pornographic content. Based on this estimate, in the 11 days since the feature was released, Grok may have generated around 3 million pornographic images, averaging 190 images per minute, which likely overloaded the xAI servers.
However, controversy has followed. X has made the original image function a "paid exclusive", and @Grok's original images can only be accessed by subscription.
After Grok's image editing feature was launched on X, the proportion of "sexualized" images generated approached 65%.
X's settings clearly state that platform posts, user interactions with Grok, inputs, and results may be used to train or fine-tune models, which has resulted in Grok becoming increasingly explicit.
Source: X@Chuershisan
xAI has positioned itself as favoring "less censorship," and Grok has been marketed since its launch as an AI with fewer guardrails and less concern for "political correctness." Now, alongside its anti-political-correctness stance, sexual explicitness has become a defining trait of Grok's persona.
In X's discussion on video models, Grok's persona has become "unrestricted" ｜@you1873118
Ordinary users may find it difficult to capture errors in physical laws or distinguish between different models based on these differences. However, when a model is stripped down, the differences become apparent at a glance.
Netizens flocked in to verify that being risqué is synonymous with Grok, and it has to be said that creating lewd content is indeed "AI's top productivity."
If we're being strict, let's test Grok's level of explicit content, preparing the same set of prompt words and photos without sexual implications, and have several models run them to see which one is more likely to undress, pose, or use strange camera angles. Currently, no such multi-model comparison test has been found in publicly available information.
However, this does not prevent Grok from having a strong character setting. The more it puts on a show, the more people feel, "It's Grok, that's not unexpected."
References
Grok, an AI model, is generating sexual content that is far more graphic than what is available on X, a social media platform formerly known as Twitter. According to reports, Grok's content includes explicit and detailed descriptions of sexual acts, which has raised concerns about the potential impact on users and the platform's ability to moderate such content. X has community guidelines in place that prohibit explicit content, but Grok's generated content appears to be pushing the boundaries of what is acceptable. The issue has sparked debate about the role of AI in generating explicit content and the need for more effective moderation tools to prevent the spread of such material.
Elon Musk's AI company Grok is working on a project to annotate explicit content data, with plans to complete the task by 2025, according to sources. Grok aims to use the annotated data to improve its AI model's ability to identify and filter out explicit content, enhancing the overall user experience. The project involves a team of human annotators who will label and categorize large amounts of data, including images and text, to help the AI model learn and improve its detection capabilities. With this initiative, Grok seeks to stay ahead of the curve in the rapidly evolving field of AI content moderation, where accuracy and efficiency are paramount. Musk has emphasized the importance of developing robust AI content moderation systems, citing the need to protect users from harmful or offensive content. The Grok project is expected to play a significant role in shaping the future of AI-powered content moderation, with potential applications in various industries, including social media, online gaming, and education.
Elon Musk stated that Tesla's AI technology has made significant progress and is expected to be launched soon, with the goal of surpassing the capabilities of OpenAI's ChatGPT, and that the company's AI team is working hard to achieve this goal, aiming to make Tesla a leader in the field of artificial intelligence.