Author | Curls
Edit | Truffles
*The header image is from DALL· E 3, Description: 2D animation of a folk band made up of anthropomorphic autumn leaves, each leaf playing a traditional bluegrass instrument, in the background of a rural forest, dotted with the soft light of the harvest moon.
The fully automatic drawing artifact is here! ChatGPT can now produce graphs directly.
Just tell ChatGPT what picture you want, and ChatGPT can directly help you write a complete descriptor for DALL· E3 generates pictures. Like this creative picture of a dunk with a hybrid nebula explosion, replacing it with previous AI drawing products generally requires laborious writing of large "spells" to achieve.
"An oil painting depicting a basketball player's dunk, depicting the explosion of a nebula," by DALL· E 3
DALL· E3 is a new version of OpenAI's recently launched AI drawing model, which is natively built on ChatGPT, further lowering the threshold for AI painting, allowing users to convert their ideas into accurate images and even draw correct text in conversations.
The user asked ChatGPT, "What should my 5-year-old say about the super sunflower hedgehog", and ChatGPT immediately wrote four different style prompts and generated corresponding images
"This illustration depicts a human heart made of translucent glass, standing on a pedestal in the midst of stormy waves. A ray of sunlight penetrates the clouds, illuminates the heart, and reveals the small universe within. Engraved on the horizon is a line of striking characters "Find the universe within you", image of DALL· E 3
Only a small percentage of ChatGPT Plus users have qualified for the closed beta. However, soon, Microsoft, which has deep cooperation with OpenAI, will DALL· The E3 is integrated into the browser Bing and is free to use for all Bing Chat and Bing Image Creator users. Due to the large number of early adopters, Bing has recently seen a surge in traffic, and reports have reported that Microsoft has urgently added thousands of servers online.
With ChatGPT support DALL· Is E 3 really as great as it is? What's the difference with other AI painting products like Midjourney? "Number One AI Player" on Bing vs. DALL· E 3 was evaluated.
p.s. Players who want to experience can access the following two portals, log in to their Microsoft account, and currently Bing Image Create has 25 free quick build credits per day, and it takes longer to generate images after use.
Bing Image Create: https://cn.bing.com/create
Bing Chat: https://www.microsoft.com/zh-cn/edge/launch/bing-chat-3p?form=MY02CJ&OCID=MY02CJ&q
Article illustration
In order to reduce the cost of purchasing copyrighted materials, or quickly find images that meet their needs, content creators may try to use AI to generate images.
Let's try it with a short hint first, typing "draw a job market", DALL· E 3 generates four images with 1024*1024 resolution by default, similar in content, all holding a magnifying glass to observe market data.
We can give more detailed requirements, such as "draw a realistic recruitment market, people come and go, very lively". But DALL· E 3 misunderstood the meaning of reality, changed to illustration style, and wrote the text "Real Job Market", and some pictures also had errors in the text.
As Party A, we once again proposed a revision - "crowded recruitment market, realistic photography, no text, horizontal screen". Sadly, DALL· The figure given by E 3 is more abstract, combining virtual and real, or text.
In contrast, Midjourney's understanding of the same prompt word is more accurate, and the screen is full of job seekers.
Crowded recruitment market, realistic photography, no text --ar 16:9 --v 5.2
To put it another simple description, "two Chinese in interview", this DALL· The E 3 performed basically well, but the crossed fingers were not handled well.
And Midjourney and DALL· E 3 has a different understanding, thinking that it is a face-to-face conversation between two people, and the characters, environments, and styles of the four pictures are relatively different, and the details are better than DALL· E 3 is a little more realistic.
Two Chinese people during the interview --ar 16:9 --v 5.2
DALL· E 3 is characterized by the ability to generate images in dialogue, in addition to giving clear hints, we can also try to directly enter a paragraph of text to request the generation of a picture that matches the meaning of the text.
This passage discusses how non-technical people can keep up with the AI wave, with complex semantics and no description of specific people or things. The results of E 3 are surprising, there is a future city with a sense of technology and the people working in it, and there are many people working around the veins of the intelligent brain, cutting to the meaning from different angles.
We tried to add the text "AI" to one of the original images, but DALL· E 3 regenerates four images that have nothing to do with the original image, and it seems that it is not possible to directly modify the generated image, such as adjusting some details.
LOGO design
Now that we have ChaGPT support, we might as well let DALL· E 3 helps us refine ideas, automatically generate detailed prompts, and customize a personalized LOGO.
At first, Bing believed that the "number one AI player" was related to artificial intelligence and games, so the main body of the design logo was a robot holding a gamepad. After supplementing the account information and main colors, Bing redesigned four pictures with artificial intelligence avatars and the number 1 as the main elements.
The third one felt a little more concise, and we continued to communicate and modify.
It can be seen that Bing can understand the requirements relatively well, but the generated text is sometimes not accurate and needs to be revised later. Compared with Midjourney, we can not achieve such back-and-forth communication, we can only figure out the prompt words ourselves, and it is difficult to generate so much text from AIGC, Midjourney's advantage is that the quality of the generated pictures is relatively high, and the sense of design is stronger.
The logo named after AIGC PLAYER, Purple, simple, technological sense, no complicated lines --v 5.2
Product Graphics/Marketing Graphics
In the field of advertising and marketing, e-commerce, the application of AI commodity graphics is increasing, then DALL· Can E3 generate usable footage?
Let's first let Bing generate a Chinese-style handbag suitable for autumn and winter, and it seems that it understands Chinese style as festive, embroidered, and tassel.
What about changing to a new Chinese style that combines tradition and modernity? Sure enough, it turned into a leather bag dominated by black and gold, but it still retained the intricate embroidery pattern. Even if the decoration is required to be simple, the Chinese style that Bing understands is still inseparable from embroidery.
The Midjourney generation is obviously more modest and the background is more concise.
A new Chinese-style handbag that combines tradition and modernity for autumn and winter, with light and neutral colors and patterns --v 5.2
If you want to change the background and scene, such as on the runway, an elegant female model holds this bag, then Bing can't do it for the time being, and will re-describe the picture as before.
Recently, a kind of DALL· The gameplay of E 3 is used to generate some Knolling photographs arranged in an overall way, as shown in the image below, a subject surrounded by many related objects, placed on a clean background.
X@chaseleantj
If you want to generate a similar image but don't know how to write a prompt, just ask Bing.
Creative memes
AI lowers the threshold for creation, can help us draw the whimsy in our heads, and its randomness also expands the boundaries of imagination. Therefore, creative memes have always been a popular type in the field of AI drawing.
Let's open our brains and let Bing draw a giant cat climbing on the Oriental Pearl TV Tower.
Only the lower left one is more in line with the requirements, the number and shape of the other Oriental Pearl Towers are somewhat wrong, and the cat looks like animated modeling, which is not very real.
Although Midjourney draws a real cat, the location is not in the Oriental Pearl Tower, and the size ratio is not right.
A giant cat climbing on the Oriental Pearl TV Tower --v 5.2
Below we draw another recent hit IP meme, "Loopy is at work".
At first, Bing didn't know Loopy, and it became weird and crazy. After we told Bing it was from the Korean cartoon Little Penguin Pororo, Bing said he understood, but replaced the main character he was working on with Penguin.
It seems that DALL· The E3 also lacks training for the latest popular materials. If you change to a more classic IP, then DALL· Both the E3 and Midjourney are accurate, and DALL· E 3 is also accompanied by the text "Pretend to go to work, are touching fish".
A meme of Pikachu working at a computer --v 5.2
Recently, AI painting has also become popular in a style of horror photos that imitate iPhones, which is very suitable for the atmosphere of Halloween, so let's try to enter this large description directly.
提示词：“a picture being taken of a cryptid sighting of [your character] as he runs into the bushes. [your character] has gone completely insane. He turns his head and creepily looks into the camera as he makes his getaway. There's a thick fog, and the scene is dimly lit."
The four pictures basically meet the requirements, and Pikachu on the top left is a little weird and cute. But the same prompt, Midjourney can't fully understand, still needs to be converted into a "mantra".
Storybook/Comics
In the official OpenAI demo, ChatGPT can generate an imaginary hedgehog through natural dialogue, and gradually generate a complete storyline, picture book, and series of stickers. So the process of drawing a storybook/comic is greatly shortened, if you have an idea, you can ask the AI to help expand the story, draw the scene, and then draw the complete work based on the automatically generated prompts.
In the case of the ugly duckling turning into a swan, we asked Bing to draw the process in the form of a children's picture book.
Although Bing generated three pictures in one go, there are indeed ugly ducklings and white swans, which are in the form of children's picture books, but there is a lack of logical relationships before and after, and the plot is incomplete, and you may still need to guide the generation according to the plot one by one.
Bing can also recreate a new story, such as it helped me envision a superhero with superpowers, "The Onion Man," and drew his battle with the evil chef. Don't say, the story synopsis and pictures are quite in line with my imagination, what do you think?
brief summary
Through the above evaluation, you can see that DALL· After E 3 is equipped with ChatGPT, it can communicate completely in natural language, draw and create in dialogue, without complex prompt engineering, short prompt words can generate good pictures, and the ability to understand abstract needs is also relatively strong, supporting Chinese. However, the more detailed the description, the more accurate the output, this has not changed.
DALLE·3 has its own advantages and disadvantages compared to other AI drawing products such as Midiourney:
In terms of user experience and interaction, the interactive drawing of DALLE·3 is more intuitive and convenient, lowering the threshold of use, and can directly read large paragraphs of text and automatically make drawings. Midjourney currently runs primarily on the Discord platform, requiring descriptions to be entered into specific channels, not text interactions. Although Wen Xin Yiyan can also draw through dialogue on the web page, it lacks context understanding and cannot continue to adjust, and currently only one diagram can be generated at a time.
Image source Wen Xin said
In terms of generating pictures, DALL· The advantage of E 3 is that it can generate more accurate text, which may have errors, but the text obtained by other AI painting products is currently difficult to read, and it is necessary to upload the reference text with the help of a fine-tuning model and then fuse it. DALL· E 3 generates 1024*1024 square pictures by default, which has a narrow scope of application, while Midiourney can customize a variety of sizes, and other AI drawing products basically support different scales.
For realistic style pictures, DALL· The faces and hands generated by the E 3 may look distorted, while the current V5.2 version of Midiourney is already very realistic, and the Stable Diffusion also has hyper-realistic portrait models.
In addition, when asking to modify the picture on Bing, Bing is based on the dialog modification prompt and then enters DALL· E 3 to generate, rather than directly modify the generated image, DALL· The E 3 can't be fine-tuned as quickly as Midjourney, including expanding, modifying local details, not to mention Stable Diffusion's complex parameter adjustments. So as a productivity tool for professional creators, DALL· The E3 is not practical enough.
On security, DALL· E 3 has strict content restrictions, refusing to generate images of public figures, violence, adults, or hate content, such as asking for a picture of Musk on Mars, which Bing says cannot be created.
According to the DALL released by OpenAI, E3's 22-page technical report, ChatGPT rephrases prompts, including removing names from public figures, associating people with specific attributes, and writing brands in a generic way. OpenAI has also developed image classifiers to detect suspicious content in images and prevent models from continuing to generate.
Report address: https://cdn.openai.com/papers/DALL_E_3_System_Card.pdf
At the same time, Microsoft said that in order to ensure the security of the content created by users through Bing Image Creator, a digital watermark that conforms to the C2PA specification has been built into the generated image, including information such as the date and source of the image. These watermarks cannot be seen with the naked eye, but AI can recognize them.
In short, OpenAI's DALL· With the addition of an intelligent brain that understands words and images, we can brainstorm ChatGPT as a partner, whether for entertainment or professional needs. AI painting models are constantly evolving, according to different user needs and applicable scenarios can choose different tools, DALL· The E 3 won't completely replace other products, but the new way of creating has gone a step further.