I give Van Gogh as Tony: These three AI drawing tools are outrageous

Source: fruit shell

This article is about 2500 words, it is recommended to read for 5 minutes

AI generates image completion, and does not mean that creativity is over.

Many people say that this year is "the first year of AI painting". First, Disco Diffusion fired, and developed communities and creative design industries from Text-to-Image (generating images), and the fire reached the vision of ordinary people.

People are enthusiastic about the two objects that do not match the world, such as "Da Vinci" and "iPhone", enter the AI ​​program, and then wait for the screen to render.

For another example, I knead the poached eggs into the clouds 丨 The author uses disco diffusion to generate

It was a "blind box" experience. For those who do not have any art foundation and painting ability, most of the "melting" diagrams of AI are amazing enough. Even if the effect "rollover" can continue to be optimized by adjusting the description.

Immediately after, Midjourney, AI painting tools, was also hot. Different from the humble interface of Disco Diffusion full -screen English and code, Midjourney is directly equipped on the Discord channel. The process of input instructions is no different from sending WeChat. What is even more surprising is that it usually takes about 60 seconds to generate paintings in about 60 seconds. Essence

God said: "There must be Wi-Fi" 丨 The author generates dall · E 2

Then, OpenAI's Dall · E 2 was killed. Unlike the previous two good at "conceptual painting style", dall · E 2 is more "realistic". You can generate 10 pictures in less than 60 seconds. Re -generating ... In just a few months, the title of "Strongest AI Painter" was easy to change.

Google couldn't sit still. At the end of May, he introduced his own player -IMAGEN, which was directly called Dall · E 2. It is known as "unprecedented realism and deep language understanding" and has not yet been opened.

In the past two months, I have frequently dealt with the previous three "AI artists". I have tested description words and tuning robots almost every day. I stepped on a lot of pits and turned over a lot of cars. But at the same time, I have gained a lot of masterpieces.

This time, I will compare their painting characteristics, user friendship, etc. At the same time, they will sort out their URLs and some simple operation methods.

With ordinary users, they are a powerful tool for imagination; in the professionals, if they link them with other tools, they can have endless imagination.

Disco

Use entrance:

https://colab.research.google.com/github/alembics/disco-diffusion/blob/main/disco_diffusion.ipynb

Disco Diffusion generates the process of generating paintings about these steps: open the program; set parameters such as picture size, process chart, and generate drawing; write the descriptor (PromPTS) in English. The format is roughly "painting type+ objects+ object (You can have multiple)+ painting style setting+ some rhetoric that plays a limited role "; then start running, waiting for AI rendering painting.

The description I wrote to AI: "A Beautiful Painting of a Starry Night, Shining Its Light Across A Sunflower Sea by James Gurney, Trending on Artstation."

Generally speaking, you need to wait for half an hour. If you stare at the screen, you will see that the image is full of noise, gradually becoming clear and detailed.

During use, Disco Diffusion may indicate that you can run enough memory over the computer, but because it runs on the computing resources such as GPUs provided by Google for free, it is not high in the user's computer hardware. Open the browser and run.

Draw a Mobis -style scene with AI: "A Beautiful Painting of a Spaceship Flying Over A Desert by Moebius, Trending on Artstation."

Disco Diffusion itself is a free open source software, but if you want a faster drawing speed, you can buy Google Colab members to allocate to faster cloud computing resources.

In addition to entering only text for AI to play freely, you can also set up a initialized image in advance to restrain the creation of AI.

For example, I first made a bottom diagram (left) with the outline of the trees and the green color block, and then operated, the Disco Diffusion will be played on this large framework.

Disco Diffusion generates theoretically commercial use. The program is based on the MIT open source protocol. All Internet users can use, copy, modify, and even sell the generating graphs for free. But I think there are risks. Risk mainly comes from your descriptive words that attract disputes over plagiarism. When you use a distinctive style of artist (especially the artists in the world), and a commercial work as a keyword, please do not directly use it.

MIDJOURNEY: Not very "super outline", more "obedient"

MIDJOURNEY is still inviting system, internal test address:

https://o9q981dirmk.typeform.com/to/zztf1mvc?typeform-source=midjourney-gallery

In order to test the production effect of Midjourney, I copied the keywords that had "feed" to Disco Diffusion- "Starry Sky", "Sunflower", "Van Gogh" -pasted in.

I use the painting generated by Midjourney

Seeing the finished product, I have an intuitive feeling: Midjourney's imagination does not have Disco Diffusion as "super outline". But if you consider the perspective of auxiliary creation, I will be more inclined to use Midjourney, a more "obedient tool", after all, no creator is willing to give the creative dominance to AI.

The advantage of Midjourney is: fast. The software generates is very fast, and one is calculated for about 60 seconds. If you are not satisfied with the finished product, you can also improve the details or extend the change almost in real time.

Generate 4 puppy police officers in one minute 丨 Use Midjourney to generate

Midjourney is on the communication software Discord. After entering "/IMAGE" in the dialog box, enter the descriptor in English, and then press the Enter key. This process is like chatting with AI.

After 60 seconds, you can receive 4 rendered pictures in the dialog box. If you are not satisfied with "Figure 1", you can click the "U1" button to increase the details and press the "V1" button to extend the change until you are satisfied.

So, I took Midjourney to generate "McDonald's in the Nineteenth Century" and "Eighteenth Century Workers":

The reason why Midjourney is a "product" Disco Diffusion, one is that its interface is more friendly, and the other is to build a creative community. Essence This is a "painting style" database with reference value, which is too suitable for "copying homework".

For example, I tried to generate the scene of "Love, Death and Robot", referring to the description of the two artists above the figure above, and then generated a satisfactory painting:

The "copying homework" has further reduced the threshold of the genetic decent work, but on the other hand, it will also lose a lot of fun. Do not let game cheats destroy a good game.

In terms of copyright, if you are a free user, the copyright of the image belongs to AI. After paying $ 30 a month, you can use the picture to commercialize it. But at the same time, if you make a profit of more than 20,000 US dollars, you need to give a 20% division of Midjourney.

Dall · e

I became "Tony" and used Dall · E 2 to send it to Van Gogh, and the application address:

labs.openai.com/waitlist

I waited for more than a month before I got the internal test of Dall · E 2. If Disco Diffusion is better at depicting atmosphere, landscape or conceptual art, then Dall · E 2 is good at realistic.

"Can Elephant Turn around?" I took this "Classic Party Demand" as an example to try the realistic ability of dall · E 2.

It turned around.

I asked netizens to play Party A and let the elephant do something else. For example, let the elephant swim in the aquarium:

Let the elephant and shark dance:

Let the elephant drive the Harley motorcycle on the road:

Let the elephant be called Cao Chong:

"Party A" has nothing to say.

It is no exaggeration to say that this is the best AI drawing tool I have experienced at present. The operation is simple enough, the completion is high, and the speed is so fast that can be used as a search engine: less than a minute generates 10 pictures (1024 × 1024), It can be infinitely extended and changes, and can even be removed from local reassembly. You can keep "hairdressing" Van Gogh.

In terms of copyright, OPENAI behind Dall · E 2 has a few strict restrictions: the picture generates copyright and eventually belongs to OpenAI; it is only for personal learning to explore and use it. A realistic face generation result will have the risk of portrait infringement.

Openai also claims that AI has forbidden AI to remember the face of celebrities and avoid the stereotypes of race and gender.

Before I waited for dall · E 2 internal test qualification, I found a "flat replacement" -Dall · E mini, which was made of DEMO made by the first generation of dall · E to generate fast generating speed, but the screen completion was not as good as Dall.· E 2.Durian sofa ｜ generate with dall · e mini, software address:

https://huggingface.co/spaces/dalle- mini/dalle- mini

Generating images, just the first step

"Can you let them move?" I looked at the paintings of AI back and began to find a way:

AI generates image completion, and does not mean that creativity is over.If you take it as one of the links and connect other creative processes, imagine space is huge.

Let me show the idea of the illustrator Nerko: he first uses Midjourney to generate the material he wants, and then assembles these parts.

@Nekroxiii

In his hands, AI is a "productive force".Selecting and synthesizing is still his full dominance.Before using Midjourney, he had drawn 15 years of illustration.

Edit: Yu Tengkai

- END -