Shulou(Shulou.com)11/24 Report--

Is there always a lack of words in the face of paintings? Now, Midjourney has launched a new image tool that can operate in reverse. As long as you enter a picture, it will give you an output prompt.

After Midjourney V5 was launched, netizens used it to generate countless amazing works.

In front of this AI painting tool, the gap between people may be reflected in prompt.

It took more than 80 hours and 900 iterations to complete the masterpiece "Space Opera House" with AI at the Colorado State Fair Art Competition last year. Ta directly refuses to share its own prompt.

Yes, now a lot of times, a perfect and professional prompt, has been worth thousands of dollars.

Midjourney seems to be well aware of our needs, and recently, it has launched a new feature, / describe, which can help us reverse prompt from images.

Now, if you have an image you want to imitate, but you don't know what kind of prompt to use to describe it, prompt will help you a lot. For example, the picture below is the "Yan Ye Zhi Ting" by Xin Haicheng.

After dragging into the image, Midjourney generates the following four kinds of prompt.

Based on these four prompt,Midjourney, four more images of similar style are generated.

It's perfect.

In addition, the description provided by Midjourney can inspire more inspiration for users. Traditional painters may suffer from blank canvas syndrome, and AI artists sometimes run out of words, making it difficult to find words to describe photos.

On Midjourney's Discord channel, you can start the text description process from "/ describe".

Midjourney provides a drag-and-drop area that allows you to drag into images.

Then press enter, and you can wait for prompt.

Four text prompts, including descriptive words, style details, and aspect ratio. The aspect ratio is similar to a computer in pixel accuracy, for example, if it is an 16:9 image, it will output 504 virtual 283.

After the four text prompts are generated, you can click the buttons directly and use them to generate images directly. Of course, you can also choose to optimize the prompts yourself before submitting.

The writer's windfall? What's even more interesting is that this new feature of Midjourney unexpectedly gives writers a hand.

If you can't find a perfect adjective to describe a picture, just upload the image to Midjourney, and brilliant sentences will pop up automatically.

And the text prompt returned by Midjourney is full of surprises. These include buzzwords such as "palewave", "cranberrycore" and "icepunk", as well as the names of various artists, photographers and authors.

When a person's name appears in the prompt, or even a clickable link, you can open a Google search, and if you click on it, you may get more sources of inspiration.

For example, upload a great picture of Ross Burgener, and Midjourney would describe it like this: "the aurora dances on the igloo lit by LED, bringing a blue hue to the color of the snow."

It recognizes the aurora borealis and gives advice in its tips-refer to artist Qlexis Gritchenko and photographer Nathan Wirth. The words it uses include "polar iceberg", "glowing sky" and "ethereal sculpture".

Netizens are overjoyed and have gone crazy. Now, netizens have gone crazy.

Some netizens have made use of the describe function to generate many beauties in white lace skirts.

Someone typed in a bunch of jewelry images.

The following prompt is obtained.

The images generated from these prompt are as follows.

Enter a two-dimensional cute girl.

The output prompt, the drawing effect is very amazing.

Enter a picture of a wildebeest with a foolish temperament.

The generated prompt is as follows.

The output prompt produces the big twist antelope, the temperament is more cold.

Enter a dog on the beach.

Prompt gave birth to four dogs buried in the pit.

Enter anime characters in tuxedos.

Generate the following prompt.

Generate four anime characters.

Is it RLHF? It can be seen that the recent progress of Midjourney can be described at full speed.

Jim Fan, a scientist at Nvidia AI, speculates that Midjourney has been conducting large-scale reinforcement learning based on human feedback ("RLHF"), and this may be the largest text-to-image reinforcement learning ever.

When users choose to enlarge the image, it is because they like it better. It would be a huge waste if this process is not used as a reward signal. This data is cheap to collect and fully meets the needs of the user base.

The more users you have, the better you can do RLHF, and then the more users you will get.

He also gave a disclaimer, saying that this is one of his brains, he does not know the people of Midjourney.

But they must have done so, because the UI format looks like a multiple-choice question.

Finally, it is important to note that, like image generation, describe features require the use of integrals, so you need to consider them carefully.

And now, the free feature of Midjourney has been completely turned off.

Reference:

Https://the-decoder.com/midjourney-new-image-tool-works-in-reverse/

Https://twitter.com/DrJimFan/status/1643279641065713665

This article comes from the official account of Wechat: Xin Zhiyuan (ID:AI_era)

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.