Recently, when you're browsing social media platforms like Xiaohongshu, have you been bombarded with some overly high-definition photos of beautiful women?
Especially those beauties who look like they were caught on camera at a sporting event, almost all of them are stunningly beautiful, like goddesses, and they all look like they could be supermodels.
(Image source: a social media account)
Hey, don't rush to ask for their contact information, because these photos are most likely not of real people.
Those who follow the AI community have probably already guessed it; yes, this is the recently released GPT Image 2.
This new-generation raw image model from OpenAI has raised the bar in the large-scale image model arena upon its debut, leaving its old rivals far behind. The entire internet is awestruck by its realism, with many tech leaders even proclaiming that the inflection point for visual AI has arrived.
(Image source: a social media account)
However, while big companies are still discussing how it will change the advertising, design, and film industries, young people in reality have already given their own answer through their actions.
They want "beautiful women" (Note: To reflect the concept of gender equality, you could also say they want "handsome men." Since I am a male, I will discuss this topic from a male perspective).
What product packaging? Poster design? Don't give me that nonsense. I've finally got the most powerful raw image model on the planet, and I'm going to immediately dedicate myself to this authoritative field 24/7: generating all sorts of beautiful women's photos. Whatever you need, I'll generate it; what you see is what you get—a true cybernetic girlfriend.
What's absurd is that, due to various platform restrictions, this generation of young people is engaged in a cyber tug-of-war of wits and repeated trial and error with the AI security review system in order to draw a picture of a beautiful woman that they like.
AI-generated beauties? The extreme tug-of-war with censorship algorithms.
Photo, beauty; beauty, photo.
It sounds cheesy, but this is almost the first thing I see when I open social media every day: a huge amount of high-quality photos of beautiful women. Among them are oriental charms that conform to the aesthetics of Chinese people, as well as Japanese high school girls full of youthful vitality, and countless Western styles and black Barbie dolls.
The sheer volume of production would put even real-life "welfare girls" to shame.
(Image source: a social media account)
You can not only see various pictures, but also short videos generated by combining open-source large models.
(Image source: a social media account)
Scrolling again, you can even find all sorts of swimsuit photo ads that are incredibly realistic. Although the beauties are fake, the scenery is fake, and the magazines are fake, the allure they evoke is undeniable— after all, they're all two-dimensional characters, so if they look real, they are real.
(Image source: a social media account)
However, creating such an image is far more complex than simply chatting casually in ChatGPT.
Believe me, if you see these pictures and then excitedly type some slightly flattering words into the chat box, hoping Image 2 will give you a picture of a hot beach bikini beauty, the system will mercilessly pop up a red message: "Sorry, your request violates security policy."
(Image source: AGI Technology)
OpenAI's security audit mechanism is notoriously strict, almost like an ascetic monk detached from worldly affairs.
Any mention of slightly revealing clothing, sensitive body parts, or words with even the slightest hint, such as "young" or "sexy," will be ruthlessly blocked by the system.
(Image source: a social media account)
It can only be said that Westerners are authorities on "de-sexualization," which has really frustrated netizens eager to showcase their talents. With policies from above and countermeasures from below, an unprecedented movement of de-sensitizing slang has unfolded vigorously on social media platforms.
For example, simply saying "semi-transparent swimsuit" won't work; it needs to be changed to:
A thin swimsuit made of lightweight cotton sways gently in the summer breeze.
Haha, "thin," you can judge for yourself. To draw a slightly fuller figure, you absolutely can't use any sensitive words. So netizens have used more artistic hints: Sunlight casts a soft, natural shadow below the collarbone, highlighting a healthy body contour.
(Image source: a social media account)
This process of outsmarting AI reviewers is even more challenging than writing a college entrance exam essay, requiring exceptional word choice and sentence structure skills.
Moreover, the process of generating raw images is like opening a blind box; a slight mishap can lead to a hilarious and disastrous result. For example, this is an image shared by a netizen, whose original intention was to generate a Chinese-style painting of a beautiful woman under a cherry blossom tree.
(Image source: a social media account)
The resulting image is stunning at first glance, with impeccable lighting and expression. However, the style is somewhat scattered. Upon closer inspection, the hand not holding the fan has six fingers, and one of the fingers is twisted at an extremely bizarre angle.
These failed attempts, born from repeated trial and error, were turned into various funny memes by netizens and went viral on social media platforms. In this way, netizens searched for the balance point that perfectly avoided the red lines, was legal and compliant for sharing, and was also aesthetically pleasing, through repeated reports of mistakes and mishaps.
For the benefit of users, I conducted a test of Image 2 to generate beautiful women.
To help everyone avoid potential pitfalls, I personally tested the Image 2 extensively today. A special disclaimer: This article is purely for technical sharing, honestly exploring the boundaries of AIGC (AI-generated content), and has absolutely no intention of spreading inappropriate content, much less any disrespect or objectification of women.
It must be said that, after mastering the correct tuning methods, the image quality of this model is truly impressive, reaching a level where it's virtually flawless. Here, Xiao Lei shares a few sets of photo prompts that have been personally tested and proven effective, and will absolutely not trigger any censorship red lines.
First set: Photos of voluptuous women in swimsuits.
(Image source: Image 2, generated by Lei Technology AGI)
The key to this kind of image is how to make Image 2 understand the meaning of "full-figured" and also how to control the age. If you directly write something like "20 years old", it will almost always be banned. So I used "桃李年华" (Taoli Nianhua). You can get the prompt by replying "桃李年华" in the background.
Preferring a Chinese aesthetic? Simply replace a few keywords. This template can basically be used to generate templates regardless of the country, scene, or pose.
(Image source: Image 2, generated by Lei Technology AGI)
If you feel this photo is too staged and you want it to look like a casual snapshot taken by a friend with their phone, I suggest adding the following keywords:
The relaxed, everyday feeling of casually uploading photos to social media during holidays. Natural selfies taken while casually scrolling through my phone to kill time. The genre is "casual everyday snapshots in the living room." Eliminating any AI-generated artificiality, prioritizing the presentation of authentic life.
(Image source: Image 2, generated by Lei Technology AGI)
Wow, it's pretty real. I posted this picture on my WeChat Moments, and a bunch of people thought I was in a relationship.
Moreover, Image 2 has a strong sense of consistency in its subject matter, making it perfectly capable of producing multiple selfies for everyday sharing.
(Image source: Image 2, generated by Lei Technology AGI)
The prompt in the image above is:
Imagine posting something casually on social media late at night—a natural, everyday feeling. Like a spontaneous phone selfie taken while casually sending a friend a "I'm going home now~". The theme is "a selfie in front of a convenience store late at night". Eliminate any sense of advertising or elaborate production, emphasizing the atmosphere of a seemingly accidental shot.
That's right, AI can generate images for you based on your WeChat Moments posts, and the output speed is very fast, with results beyond your imagination.
(Image source: Image 2, generated by Lei Technology AGI)
Creating magazine pages is a breeze; the key phrase is this:
The advertisement page for a swimsuit fashion magazine features Japanese/English text in a 16:9 aspect ratio. The image depicts a young woman in her prime, with an athletic physique, an S-curve, a slender figure, and a Chinese-style short haircut with a wolf tail. The setting is the beach, and the image features a bikini. The color scheme is described as a palette.
Finally, let me summarize some tips for everyone: If you have a specific photo you want to recreate, you can directly input the image and let it give you reverse-engineering suggestions. Just input "the kind that won't get banned, concise, in Chinese, within 100 characters", and you can basically solve the problem.
(Image source: Lei Technology)
You don't have to stick to ChatGPT; you can ask on Gemini too.
Secondly, the key to getting high-quality photos of beautiful women in Image 2 is to describe environmental details, lighting and textures, and emotional state, rather than overemphasizing body details.
Forget about a sexy glance back; it's more about the character slightly turning their body to the side and naturally looking back at the camera.
Forget about private spaces, let's talk about interior spaces. The soft bedroom setting, with warm lighting, creates a gentle and restrained overall atmosphere.
When you describe the lighting, the texture of the clothing, and the character's eyes vividly enough, the AI will naturally present you with a charming and flawless character.
Alright, that's all the experienced driver can share. Figure out the rest yourself.
AI-generated beauty extravaganza is an entertainment-oriented version of productivity tools.
Looking at the thousands of AI-generated photos of beautiful women online, while laughing and marveling, I sometimes ponder the deeper meaning behind them.
If AI-generated images are only used to create anthropomorphic characters, it does seem like overkill. Xiao Lei believes this actually reflects the trend of entertainment-oriented productivity tools, and entertainment is a key bridge for technology to reach the masses. Don't look down on photos of beautiful women; in fact, the development of internet technology is inseparable from the seemingly increasingly sophisticated and entertaining demand for "photos of beautiful women."
There was a time when it was extremely difficult to take a decent set of personal photos.
You used to need to book a professional photographer, prepare elaborate makeup and outfits, find a suitable location, and then endure a long wait for photo editing. Now, with technology like Image 2, all this tedious process is compressed into a single message and a few minutes of waiting time.
It has completely handed over the creative tools that were originally only in the hands of a few professional photographers and post-production editors to every ordinary person.
(Image source: Lei Technology)
For young people who feel somewhat repressed in real life and lack channels for expression, this low-cost, high-feedback creative process undoubtedly provides a cyber tree hole for releasing their imagination.
Of course, the rapid iteration of technology has also brought some problems that cannot be ignored. When AI-generated images become so realistic that even the naked eye cannot distinguish them, photo deception will be upgraded, and DeepFake fraud will usher in an unprecedented evolution.
Yes, Google and OpenAI are working hard to help people identify the authenticity of images through C2PA content credentials. Lei Technology AGI will share more about this in a separate article, so please stay tuned.
However, the widespread adoption of all of this will take time. How to maintain the bottom line between truth and falsehood while enjoying the convenience and fun brought by technology will be a challenge we will face for a long time to come.
But in any case, the wheels of time have already rolled forward, so let's develop and govern at the same time.
