I used AI to generate an advertisement in five minutes, but it took me five hours to "get rid of the AI flavor"
Since the end of last year, more people have felt the "invasion" of AI into their lives - we have begun to see AI-generated advertising images in subways, elevators and stores.
The reason why it is called "invasion" is because these images not only have a strong "AI flavor", but also have uncorrected irregularities, causing many people to shout "hell".
This kind of advertising is simply spending money to tell people not to trust this brand - even the advertising is so casual, how can the product be trusted?
However, many advertising agencies seem convinced that AI is the future.
The “AI flavor” that offends consumers has not reduced advertising companies’ interest in AI, but has instead sparked a round of workflow innovation that “removes the AI flavor”.
When we are “getting rid of the AI flavor”, what are we getting rid of?
Pictures come from: 小红书@rataalada, @猫头哥
At present, in the midst of the rapid development of AI, "AI flavor" is destined to be a word that will continue to be redefined.
Nowadays, "AI flavor" in images mostly refers to a certain kind of "high-definition and blurry" - bright and smooth, lacking real details.
As humans, our eyes are drawn to imperfect details.
So when we see these (AI-generated) images now... I find it quite disgusting, to be honest.
said industrial designer Ti Chang.
Because of this, how to find those "imperfect" details that can "turn the fake into real" has become a major focus of "removing the AI flavor".
You may be able to use AI prompt words to generate them at first, but then you have to adjust them, and sometimes you even need to use traditional tools like Photoshop to make it better.
Visual artist and designer Vaunn Yevo shares.
When making business orders, Yevo will first use AI to generate images, and then use PS to add "imperfections" to the portrait, such as pores, fine lines, or tiny hairs on the face.
Yevo also tried to use AI prompt words to directly add these "imperfections" to the image, but the effect was not ideal: "Sometimes if you ask the AI model to add freckles, the effect will make the entire skin look very... Strange."
The picture on the left is Yevo’s signature style, and the picture on the right is a photo generated to imitate his style as much as possible. The picture on the right also highlights the difficulty of using AI to generate natural portraits: messy fingers are fatal, and extremely exaggerated freckles are also very unnatural.
There is no doubt that when someone finds a rule that can "turn false into real", someone will start to make AI tools based on this.
At the end of 2023, a tool called “Magnific” became popular on the Internet.
This tool, called "AI Image Refinement Master", can add more details to AI-generated images and supports different generation modes. It has attracted 400,000 registered users in just over a month after its release.
The picture on the right shows the effect after Magnific retouching
McCann Worldgroup, a world-renowned advertising company, is now trying to add tools like Magnific to its workflow.
In the diagram made by McCann Worldgroup for the cultural organization Black & Abroad, the originally blurry AI-generated diagram has been greatly improved after detailed adjustments (right picture).
From the comparison, it can be seen that the second one not only adds more skin and hair details to the portrait, but also adjusts the background scene and the drink in the character's hand.
At the same time, McCann Worldgroup is also trying another method of "using AI to treat AI" - a multi-model + artificial hybrid model.
When designing a digital ad for Mexican food company Bimbo last year, McCann Worldgroup used one model to generate the foreground of the image, then another model to generate the burgers and hot dogs in the image, and another model to generate the background. As for the poster, For the font part, manual design is used directly.
It's a back-and-forth conversation with technology and technical experts, but we have a unified but evolving creative vision that has many layers to it.
explains Ian Mackenzie, global AI creative lead and Canada chief creative director at McCann Worldgroup.
As mentioned earlier, we have a specific style definition of “AI flavor” in this period.
An advertisement at this year’s Super Bowl was criticized for being too “AI-flavored”, but the photos were indeed taken by human photographers
This also means that as long as you step out of that style, people will not feel that the "AI flavor" is so strong, even though it is actually an image generated by AI.
Digital marketing company Media.Monks has assembled a special team of art creators this year: animators, color experts, directors, photographers, etc. The special thing is that they not only understand AI image generation technology, but also can create unique visuals. aesthetics.
With the collaboration of this team, Media.Monks made a very detailed design of the AI prompt words to more fully express the color, lighting, saturation, photography angle, depth of field, etc. of the desired artistic style, and finally generated and Generally "AI flavor" quite different stylized images.
The production company Tool and the advertising company Ogilvy have directly built their own specialized design systems. They can use the past materials of the service brand as materials, transcribe specific parameters for the system, and finally create generated images that are consistent with the brand aesthetics.
Antonis Kocheilas, chief transformation officer of Ogilvy, said that under this method, the occurrence of "six fingers" generated by AI will be reduced, but it is still not completely avoidable.
I don't think generating "dummy" is necessary right now.
said Dustin Callif, President of Production at Tool. In his opinion, it would be more appropriate to use AI to generate surreal and more fantasy images.
The AI-generated picture theme made by Tool is more surreal
Why is “AI flavor” not welcomed?
Our dislike of “AI flavor” sometimes comes not only from the exhaustion of style and aesthetics, but also from our resistance to the metaphor behind it.
A reader once wrote to Wired to share an interesting story:
I have an artist friend who gave me an AI-generated painting as a gift.
I can see that she was trying to make the concept more personal, and it was framed beautifully, but I still feel like I was ripped off.
In a reply, Wired affirmed the reader's sentiments, noting that the images generated by the model neither "require any financial sacrifice" nor "require any real creative investment beyond writing the prompts." ”, which “lacks the unique imprint of your friend’s creative mind.”
To put it bluntly, it means "not sincere" and "not sincere".
This aversion to “AI flavor” is not even limited to images.
Paul Graham, the co-founder of Y Combinator, once complained in a post on A very commonly used word for models.
Under this post, Ankita Gupta, founder of the AI company Akto, also said that she deeply resonated with it. She also started to turn around and leave when she saw those "iconic ChatGPT words":
There is nothing wrong with these words, but they make human language more mechanical.
Obviously, this sentiment is not going to be around for a long time. After all, there are many "Humanizers" in the writing tool recommendations on the GPT store now, which are tools that make AI-generated text look less like AI text.
What’s even more exaggerated is that there are even companies now promoting “100% no need for large language models” as a selling point.
This can also be regarded as the ultimate way to "remove the AI flavor".
Returning to the field of AI images, the disgust that the “AI flavor” makes us also comes from the fact that it takes away a special emotion-the sense of awe.
Because AI is so "omnipotent", now when we see a photo that used to make you say "Nature/the world is so amazing", the first sentence that comes to our mind has become ——"Is this generated by AI?"
Photographer Aytek Çetin’s work is now often suspected of being AI-generated images
Prior to this, social media and short videos had already raised our "stimulation threshold" extremely high, while AI imaging technology was almost filling our minds, making most things cheap and boring.
However, we need to feel surprise and awe.
It’s one of those fundamental feelings that makes us human. "Big Think" author Kevin Dickinson explained in an article exploring "awe" from a psychological perspective:
Awe is the process by which people begin to reevaluate their thoughts and beliefs when encountering new experiences or information.
In other words, when we experience a sense of awe, people begin to question their worldview and may move toward change.
The power of nature, or the beauty of human achievement, are things that reduce our self-centered influence and require us to reexamine our understanding of the world and our place in it.
It feels like a loss.
But it’s not like we haven’t felt this loss before.
The popularity of Photoshop and other image editing software has made the saying "where there is a picture, there is the truth" obsolete.
As digital image processing technology developed, people began to panic, thinking that photography was dead. But that's not true.
The media has always been "manipulated", often used to create sophisticated "deceptions."
Derek Conrad Murray, who studies art history and visual culture, commented.
As AI technology develops, the “AI flavor” will continue to change and is likely to become increasingly subtle and difficult to identify.
In this new Internet world full of uncertainty, Jason Parham, the author of "Wired", has some practical advice:
Let us embrace being distorted and accept living in the impermanence of images that are full of deception.
We need to remain diligent, because the future is a playground of constant understanding and “understanding,” of collapse and rebuilding.
This article comes from the WeChat public account , author: Fang Jiawen, 36 Krypton is published with authorization.