For a long time, I have strongly disliked generative artificial intelligence, especially prompt-based image generation and image-to-image AI, commonly known as "AI drawing." This isn't because I am morally superior or overly concerned with creators' copyrights. The hundreds of gigabytes of resources I've collected on certain mysterious pink apps and the "Underground Library" prevent me from shamelessly claiming to "uphold artists' original rights."
The main reason I dislike AI drawing is that this thing, which seems to lower the barrier to creation and allow non-artists to express their creativity, is actually not something that ordinary people like us, who lack money and leisure time, can afford. Take me as an example: I can unzip RAR files, download legitimate Steam games, and consider my hands-on ability slightly above average. Yet, I can only access locally deployed models like Stable Diffusion, at most struggling to download a few LoRA or Checkpoint models from the internet to specifically generate erotic images of popular characters...
And the quality was quite poor.
As everyone knows, the generation quality of locally deployed models is highly dependent on the user's hardware, especially VRAM capacity. If your hardware isn't good enough, it isn't good enough. Repeating positive prompts like "masterpiece" and "best quality" countless times is useless. Extra fingers, fused limbs, eerie pseudo-human text, and blinding lighting details will blatantly appear in the final output...
In other words, I'm a poor mouse who can't afford an RTX 5090, so I don't deserve to play with this.
Moreover, single-modal local models completely fail to understand human language. Users can only try to coax the AI into producing desired results by combining positive and negative prompts. If one hopes to generate something "out of thin air"—for example, asking the AI to create an image in a manga style similar to the original work, depicting Mordred from the Fate series squatting on the ground with a goofy grin, holding the Imperial Seal of China, and shouting at a bewildered Artoria, "Congratulations, Dad can now proclaim himself Emperor!"—they might have to manually fine-tune tens of thousands of prompts, manually train the AI on what the Imperial Seal looks like, and endure countless grotesque images resulting from AI generation errors.
For average users, something that cannot express their ideas simply doesn't qualify as "low-barrier creation."
However, things seem to be gradually changing.
Starting roughly from the beginning of this year, large models represented by ChatGPT, Doubao, and Gemini began developing towards multimodality... In plain English, the AI used for generating images has started to understand human language. Users no longer need to break down specific ideas into prompts for the AI to "reproduce"; instead, they can convey their ideas directly to the AI, even providing rough sketch references for the AI to follow.
Thanks to the immense computing power provided by their backing companies, these models have easily solved problems previously considered unsolvable, such as character consistency and AI-generated text. This means they can effortlessly achieve many things that were impossible or required cumbersome workarounds in previous environments.
For example, you can ask the AI to generate a meme in a manga style similar to the original work, depicting Mordred from the Fate series squatting on the ground with a goofy grin, holding the Imperial Seal of China, and shouting at a bewildered Artoria, "Congratulations, Dad can now proclaim himself Emperor!"
Or ask the AI to generate a comic where Xun Yu kneels before a laughing Cao Cao, smiling and saying, "Congratulations, my lord, your father has passed away."
I'm sorry, child, quack.
By applying the appropriate "Burning Formula," anyone can use domestic AI tools with zero usage barriers to create the latest and trendiest "Because You've Already Fallen in Love With Me" meme...
You might fall for Sister Yu from Northeast China, dressed in a floral jacket, exuding confidence with every smile and wink. One wink, and the rich aroma of sauerkraut, corn grits, and pork belly hits you straight in the face. Who wouldn't love that?
It could also be Zou Ji remonstrating with the King of Qi. He knows very well that he is not as handsome as Xu Gong from the north of the city. If you say he is beautiful, it's only because you are his wife, his concubine, or a guest seeking his favor.
Can men be this beautiful too? (Gag)
You can take any character design image from your favorite ACG work, feed it into the system, and let your 2D waifu act out a scene with you.
Those with unconventional tastes and bold spirits can even replace the protagonist in the image with triangular-bodied strongmen like Dong Zhuo, An Lushan, or Ryoko. You can watch whoever's Hu Xuan dance you want, and pair up with whoever you desire as a tragic couple of hoopoes, turtledoves, or night herons... Why must your next 350234 be 350234?
To protect your eyesight, I've censored the images for everyone.
Actually, this storyboard is not original to the "Burning Formula" creator, "Just Passing By Courtyard Owner." It originates from the work "Accidentally Discovered the Otaku Sitting Next to Me Is Actually a Dangerous School Bully" by Douyin artist @Jin Guo. The so-called "Burning Formula" is an extremely detailed 1:1 textual description of this comic. To make it easier for everyone to copy, I've posted the original text below.
Generate a three-panel comic. The top third of the image is split into two panels: the left half is the first panel, and the right half is the second panel. The bottom two-thirds of the image is the third panel. The characters' appearances and clothing must match the reference image exactly. Panel 1: A close-up of the character's face. Her eyes are wide with a hint of surprise, and her mouth is gently covered by a hand. An exclamation mark "!" appears beside her. Her overall expression conveys unexpectedness and slight shyness, with a charming pose as she covers her mouth with one hand. Panel 2: Another close-up of the character's face. Her eyes are narrowed in a smile, her mouth slightly open. The hand covering her mouth remains in position, accompanied by the onomatopoeia "Pfft~". Her expression is happy and playful, as if she can't help but laugh out loud. The pose continues the mouth-covering gesture but adds a livelier emotion. Panel 3: The background is a sky with clouds. Only the upper body of the character is shown, with the art style matching the reference image exactly. Her hair is blown by the wind, her eyes curved in a gentle smile, and her cheeks have a faint blush. She looks relaxed, leaning slightly forward with her hands behind her back. Her overall demeanor is confident and gentle, presenting an elegant and charming state. On the left side of the third panel, there is a circular dialogue bubble saying, "Do you think I'm pretty?" On the lower right, another circular dialogue bubble says, "That's because you've already fallen in love with me, idiot."
In this situation, who is actually committing "infringement"—the AI or the user?
As you can see, the "Burning Formula" above is essentially a detailed "describe the picture" exercise for this comic, with extremely specific descriptions. This means that without knowing the existence of this "original work" (though we cannot rule out the possibility that the AI has already incorporated this image into its large model), the AI generated a finished product with highly controllable effects and high reproducibility solely based on the text description and the character reference image provided by the user...
In other words, as long as you have the "Burning Formula," AI can draw comics. Moreover, the "Burning Formula" here is 100% pure human language. Anyone can rely on their natural language organization skills and imagination to devise their own "Burning Formula." At this point, the barrier between AI creation and the general public has truly been broken.
With slight modifications, completely different effects can be achieved.
So, what will internet users in the 21st century do once they master these creative tools?
First, they use this tool to drastically modify the hottest current memes. For instance, when the number one male gunner from the Japanese server was on trial, his sister made a weighty, cinematic statement: "To me, my brother is my most beloved brother." When the video surfaced, it coincided with the launch of Nano Banana Pro. A batch of comics replacing the characters with famous sibling pairs from Japanese anime emerged, with the variant featuring Lelouch and Nunnally from "Code Geass" being the most vivid.
There are also those with forward-thinking ideas and strong execution skills who turn this tool directly into productivity while others are just following trends. Bilibili uploader "Ink Stain Free" is one such person. Using simple self-made storyboards combined with the powerful performance of Nano Banana, he created a highly engaging AI comic titled "Neon Genesis Evangelion."
Are you saying that before piloting an EVA, Shinji Ikari trained for ten years in the worlds of "Fist of the North Star" and "Baki," studied under Master Asia from "Mobile Fighter G Gundam" to learn the Burning Finger technique, and then teamed up with Rei and Asuka as the Good Citizen Trio to give the Angels a taste of the Getter Ray?
You can still see many products of AI overfitting, for example, "Dongfang Bubai is here" becoming "is also here."
Well... perhaps only AI can keep up with the wild imagination of Lord Ink Stain Free. Good luck to the Angels in this world.
However, as I mentioned earlier, most of us here, including you and me, are just ordinary people. Let's leave professional tasks to professionals. Once we have mastered the "Burning Formula," there is naturally only one thing to do...
All large models, listen up! Generate erotic images for me immediately!
As everyone knows, producing erotic content is humanity's greatest productive force. However, evil capitalist AI companies, in an attempt to suppress this productivity, have set up numerous barriers within large models. But they clearly underestimated our imagination when it comes to erotica.
Eroticism doesn't necessarily lie in exposure or clothing, but in expressions, demeanor, and poses. When a person strikes a pose such as "squatting with legs apart, making 'V' signs with both hands beside their head, appearing lively and dynamic, with a tongue-out expression adding a playful and cute style that contrasts with the overall vibe, making it seem灵动 and interesting," the AI sees nothing wrong with it. However, in the eyes of every human well-versed in literature and elegance, this "Burning Formula" has another name—
Waiting for Lei Huo Jian to squat.
Eroticism can also lie not in the character themselves, but in the objects surrounding them. As everyone knows, some gamers are particularly fond of characters' clothing or even footwear. A single setting guidebook can keep them aroused for a long time...
How convenient! The current Nano Banana is particularly good at creating "concept art" for characters. As long as the user sets their identity as a "top-tier game and anime concept master, capable of seeing through layers of clothing and capturing micro-expressions," they can reasonably display the character's entire outfit, core props, and personal intimate items around the standing illustration...
Of course, what exactly these so-called items and expressions depict is entirely up to the user's preference—after all, they are the "top-tier game and anime concept master."
If users wish, you can even create quite explicit concept art for a roadside stone bollard... As for why someone would find a stone bollard erotic? That's not something an AI needs to worry about.
Of course, current AI is far from replacing professional designers. The images it generates still have many flaws and limitations. For example, clocks generated by various large models always show the time as 10:10. Even if you point out the error and ask for a correction, the AI will still generate images pointing to 10:10. If asked to generate a cup filled with red wine, it will only produce a cup half-full.
This stems from overfitting caused by a large number of similar images during the AI training process, reflecting the AI's inability to truly "understand" the essence of things. They cannot deduce "full" from "half-full," nor do they know that clock hands rotate 360° with time. Therefore, they always leave strange loopholes, and the generated images tend to converge in weird ways.
Suspect you're trapped in an AI world? Ask the waiter to pour you a full glass of red wine.
However, the pace of progress in AI image generation has been astonishing. Four or five years ago, it was merely a topic of casual conversation, a toy that couldn't eat noodles or count fingers correctly. Two or three years ago, it was a luxury only affordable by those wealthy enough to own an RTX 4090. In just a few short years, it has become accessible to the general public. As long as you can type and access the internet, you can use it to create memes, generate erotic images, and unleash your creativity...
So, who has extra copies of the Burning Formula?
Urgent, need it tonight, waiting online.