AI is playing a very new art, 7 million netizens are watching online, ControlNet has made another contribution
The dreamy west wind comes from the Aofei Temple
Qubit | Official account QbitAI
AI is playing a very new art again.
A set of pictures that "repeatedly test the boundaries between outrageous and reasonable" have swept all major platforms. The most popular one???? has nearly 7 million views and 168,000 likes . People everywhere are asking for tutorials.
In addition to the checkerboard style, there is also a spiral style that is also very popular.
Even Paul Graham , the founder of the well-known investment institution Y Combinator, came to watch:
At this moment, AI-generated art passed my Turing test.
Many netizens even mentioned in the discussion: These AI works give people a feeling of "Escher," the master of optical illusion graphics .
It also further proves that Escher, who lived in the last century, was able to draw this effect without relying on AI, and he is worthy of being a true master.
However, the advantage of AI is that it can not only draw still pictures, but can even use this style on videos to maximize the illusion effect.
(The online trial address and operation tutorial are at the end)
Everything can be delusional
This checkerboard style and spiral wind came from Reddit. It was originally a spiral map of a medieval town posted by a netizen named MrUgleh .
Then he reissued a set of works from the same period on ????:
Netizens were immediately shocked after reading it, and their thoughts surged instantly, which spawned a large number of "derivatives of famous paintings".
Ancient Greek scenes, Impressionism, oil paintings, gouache, and Cézanne styles are all available.
Whether it is buildings, rural areas, mountains and rivers, the pictures drawn by AI are all unique.
The spiral wind was dizzy, and then came the Kangkang chess and card style, which also had nine-square and sixteen-square grids.
In addition to this kind of chess and card style pictures that can clearly see the structure, there are also works by netizens. At first glance, there is almost no trace of chess and cards:
This style reminds netizens of an iceberg photo taken by Canadian photographer David Burdeny in Antarctica in 2007:
△ Photographed by David Burdeny
In addition to landscape pictures, making character pictures is also easy:
This is not over yet. A netizen named "charlesai" used the same tool to create "Artistic Gaze" based on inspiration from chess and cards and spirals:
Netizens called:
The spiral is just the beginning.
It is indeed just the beginning, and some netizens with great imaginations have come up with new tricks. Let’s just talk about using ChatGPT to generate code to make jigsaw puzzles . Who else:
After you’ve finished exploring the pictures, you can’t miss out on the videos. Well, let’s find out more about the abandoned basement:
Can the pizza video below be used directly as an advertisement?
There are also indispensable operations for converting 2D images into 3D:
Create a picture in just ten seconds and you can learn it in no time
If you also want to try this kind of gameplay, some enthusiastic netizens have already created a demo on Hugging Face.
Just open it in your browser and play.
The principle is very simple, it is the open source AI Wenshengtu Stable Diffusion combined with ControlNet (yes, Controlnet has made a contribution again).
It even used one of the same models of the last popular AI-generated creative QR code, QR Code Monster .
If you want to deploy it locally, just download this.
The online demo has prepared three basemap templates in advance: four-square chessboard, 9-square chessboard and spiral.
Next, enter the prompt word, adjust the intensity of the base image, and generate it directly.
According to the actual measurement, the default strength of 0.8 for the spiral base map is sufficient. The strength of the chessboard base map needs to be increased, and a value between 1.5-2 is more suitable.
More advanced players can also click on the advanced options to adjust more detailed parameters.
As of press time, the number of people queuing up is not large, and the picture can be published in a few seconds to more than ten seconds .
Use the chessboard base map, 1.6 intensity, and do not adjust other parameters. The effect is as follows:
Romantic painting of a ship sailing in a stormy sea, with dramatic lighting and powerful waves
If you want a new style, you can also make your own black and white base map and upload it.
Like the light and shadow character model that has been popular for a while , complex Chinese character AI can also find ways to draw it for you~
Online trial:
https://huggingface.co/spaces/AP123/IllusionDiffusion
QR Code Monster model download:
https://huggingface.co/monster-labs/control_v1p_sd15_qrcode_monster
Reference links:
[1]
https://x.com/dr_cintas/status/1703392485702086850
[2]
https://x.com/paulg/status/1701184470013894783
[3]
https://x.com/osanseviero/status/ 1703748476259950656
[4]
https://x.com/StelfieTT/status/1702728188365812071
-over-
The most “in” big model | Column article
How to compress hundreds of millions of parameters with one click? How to balance the performance and security of large models? How can one line of code optimize the model and accelerate it? Here are all the ways to play with large models, and we will teach you step by step how to speed up large model reasoning!
Click on the image below to jump to the column article page.