Sun Wu plays Warcraft? There is a picture and a truth

Alex From the Aofei temple
qubits | official account QbitAI

lately , Foreigners with big brain holes use DALL·E and Imagen Yes “ Combination of Chinese and Western ” It was a lot of fun .

Following “ Tiger wear VR” after , Sun Wu, an ancient military strategist in China, has become a new target of consolidation .

The tester gives DALL·E There is a problem ：

Let Sunwu play 《 World of warcraft II》 game .

Who would have thought , A.D. 2022 year , The grandson of the spring and Autumn period played on the computer 《 World of warcraft II》.

The whole picture is actually very harmonious , Also quite Oriental flavor .

It should be because DALL·E I entered the determiner in the token ：Oriental Painting（ Oriental painting ）, Determined the painting style .

however , How can we tell that what he is playing is 《 World of warcraft II》？ I'm afraid you still need to use your imagination .（ Manual formation ）

For the same paragraph of text entered ,DALL·E Other versions are also provided .

In this picture , It also reflects the characteristics of Sun Tzu as a military strategist .

Sitting in front of a computer and playing games may not be exciting enough , Try setting Sunwu as a character in worldofwarcraft ：

DALL·E Not only is it done , The image is also 3D Rendering .（ Although some pictures do not have the interface of Warcraft game ）

Netizens exclaimed ： This AI I actually know what the characters in worldofwarcraft look like ！

Let's talk about this “AI painter ” How to create .

DALL·E working principle

DALL·E In essence, it is a being 120 100 million parameters have been retrained GPT-3.

Its The core The module is OpenAI Of CLIP.

This is one in a variety of images 、 Neural network trained on text , Responsible for scoring and reordering the input images .

Actually ,DALL·E stay “ A literary creation ” A lot of pictures are generated in the process . In these images , The higher the matching degree with the text ,CLIP The higher the score you give .

as for CLIP How to analyze the picture and text matching degree ？ This is inseparable from its pictures and words “ The fusion ” The ability to understand .

And this ability is due to its multimodal neurons , It has a working mechanism similar to that of the human brain ： Be able to respond to the same meaning in words and images at the same time .

Final , The top rated pictures will be presented as output .（ This explains why entering the same text can generate multiple images ）

It was suggested that , May as well CLIP and GAN Combine , To achieve greater functionality ：

Give Way CLIP Calculate the similarity score between image and text description , Then give feedback to GAN, Give Way GAN To improve the score as the goal of continuous iteration .

The two generation DALL·E Combined with the CLIP And diffusion model .

among , The diffusion model sacrifices diversity , It can greatly improve the fidelity of the generated image .

CLIP Text embedding will first be transmitted to natural regression or extended prior , To produce an image embedding .

then , This embedding will be used to adjust the expansion encoder , To generate the final image . in other words , Image generation is in “ Spread ” Completed in the process .

Compared with the early generation ,DALL·E 2 Less time to generate images , Higher image resolution .

Besides ,DALL·E 2 stay “ mapping ” Shadows are also considered in the process 、 Reflection 、 Changes in texture and other factors .

for example ： On the left “1” Put a sofa ,DALL·E 2 Will analyze the light direction according to the existing information , And draw shadows .

DALL·E Of “ Weirdo ” Appreciation of works

DALL·E 2 The official in the Instagram There are many programmers and engineers on display “ Weirdo ” portrait .

There is an ancient Roman version of spider man .（ The second one is really amazing ）

Spider-Man from Ancient Rome

Picture of Jesus riding a dinosaur .（ When Jesus saw it, he would call out to the good guy ）

Jesus Christ wielding a samurai sword and riding on the back of a velociraptor, painting

Small animals wear fur 、 Wearing sunglasses 4K High definition pictures .（ These days , Seeing animals, they all feel pretty ）

A photo of a cool wearing sunglasses and a leather jacket, 4k

A sports cat explained its latest scandal photos to reporters at a news conference .

Surrealist works .

Remembrance of nostalgia, surrealist painting

And the cover of cyberpunk love story .（ Is the content of the novel OK direct Give Way GPT-3 Yes ？）

The cover of a cyberpunk romance novel

besides , There is also the use of word differences , Yes iPhone Retro dynamic graph .

One More Thing： use DALL·E Change the trademark

In addition to using DALL·E 2 Outside painting , Another researcher Janelle Shane Use this tool to try to design new trademarks for major companies .

Let's see if there's anything you like ——

DALL·E 2 Pizza Hut designed LOGO Long like this ：

Here is DALL·E 2 Designed for burger king LOGO：

And for NASA The design of the LOGO：

at present DALL·E Also in a large number of data test training , And there are certain moral challenges and other factors , It is not officially open to the public .

Official expression , Interested in trying , You can sign up and wait for the invitation .

But you can go DALL·E The official website selects the given keywords to play .

for example , choice ：“ An astronaut ”、“ Lying in a tropical resort in space ”, Then determine the painting style ：“ lifelike ”.

DALL·E You can draw several works according to the proposition .

Don't say , There are “ Space Resort ” That smell .

If you put your ideas into a picture , How are you going to create ？

DALL·E 2 Official website address ：
https://openai.com/dall-e-2/
Reference link ：
https://arxiv.org/abs/2204.06125
https://www.aiweirdness.com/ai-versus-your-corporate-logo/