{"id":42812,"date":"2025-09-11T15:07:32","date_gmt":"2025-09-11T07:07:32","guid":{"rendered":"https:\/\/www.1ai.net\/?p=42812"},"modified":"2025-09-11T15:07:32","modified_gmt":"2025-09-11T07:07:32","slug":"%e5%ae%9e%e6%b5%8b%e5%8d%b3%e6%a2%a6ai%e6%96%87%e7%94%9f%e5%9b%be%e5%8a%9f%e8%83%bd%ef%bc%8c%e7%8e%a9%e8%bd%acai%e8%a7%92%e8%89%b2%e4%b8%80%e8%87%b4%e6%80%a7%e3%80%81%e4%ba%ba%e5%83%8f%e7%9c%9f","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/42812.html","title":{"rendered":"IT'S A DREAM-A-VISIM FUNCTION, AND IT'S A GAME OF AI CHARACTER CONSISTENCY, HUMAN INTEGRITY, POSTER GENERATION, ETC"},"content":{"rendered":"

This is the real storyDream<\/a>4.0 Image generation function, four parts, respectively:<\/p>\n

1: The grouping generation function, preserving the unity of the image<\/strong><\/p>\n

2: Graphics information + serial scenario generation<\/strong><\/p>\n

2: The authenticity of the images generated<\/strong><\/p>\n

3: Poster generation and accuracy of text<\/strong><\/p>\n

Savings small operations:<\/strong>I FOUND THAT IF A DIFFERENT POSTER OR IMAGE, STYLE AND COLOUR DISTRIBUTION OR FACE WERE GENERATED IN THE SAME BATCH, BECAUSE 4.0 NOW REQUIRES A SCORE (NOT KNOWN AS THE PUBLIC VERSION) TO PRODUCE A MAP, AND A DIRECT HINT TO GENERATE AN IMAGE REQUIRES FOUR POINTS, WHICH IS QUICKLY CONSUMED, SO I ASKED AI TO PRODUCE FOUR DIFFERENT IMAGES IN THE HINT, OR JUST TWO, WHICH IS A HINT = A PICTURE, WHICH CONTROLS THE COST OF THE SCALE\/REDUCTION OF THE DRAW, HIP-HOP, WHICH IS THE REVERSE OF THE MULTIGRAPH GENERATION\u3002<\/p>\n

1. Grouping generation function, preserving unity of image<\/strong><\/p>\n

In my last assessment, I found that the image editing model of Dream 4.0 was consistent and multi-graph + text was able to generate group chart functions\u3002<\/p>\n

I'm thinking that since you're 4.0 for multi-modulars, the same model can be achieved\u00a0Ventura, image editing, group drawings\u3002<\/strong><\/p>\n

Is the image generated as consistent as the image editor? Can it be maintained in the same batch\u00a0<\/strong>Group Chart\u00a0<\/strong>Woolen cloth?<\/strong><\/p>\n

And the Nano Banana model is a one-time, one-time, one-time, one-time, one-time, one-time, one-time, one-time, one-time, one-on-one<\/p>\n

So I measured a couple of waves, and the results were real, and the graphics were super real, and the same graphics were very, very good, and the image was good\u3002<\/strong><\/p>\n

here's a two-k map, and you can take a closer look at how it works:<\/strong><\/p>\n

OneY2K SWEETY-SWEETY<\/strong><\/p>\n

CONSISTENCY OF REQUIREMENTS IN FIGURE 1: YOUNG WOMEN, Y2K SWEETHEARTS DRESSED AND DRESSED, STREET SHOOTS, ULTRA-BLANK PHOTOGRAPHY, BLUE SKY BACKGROUND, BRIGHT SUNLIGHT, PEOPLE IN FRONT OF WHITE FERRIS WHEEL WITH PINK SKATEBOARDS AND DOLLS<\/strong><\/p>\n

Tips:<\/p>\n

THE GENERATION OF FOUR GROUP IMAGE SERIES REQUIRES CONSISTENCY OF CHARACTER CHARACTERISTICS (BLACK LONG HAIR, FINE MAKE-UP, YOUNG ASIAN WOMEN), CLOTHING (IN PINK TWEEDS, PINK SHORT SKIRTS, WHITE STOCKINGS AND ARMCASES, FEET ON PINK SANDALS, OVERALL Y2K-SWEETY-SWEETY-SWEETY, BLUE SKY) BACKGROUND (LARGE WHITE SKYSCRAPER) AND IMAGE AESTHETICS (FASHION STREET, REAL SENSE PHOTOGRAPHS, Y2K AESTHETICS, ULTRA-BLANK ANGLES, LOW ANGLES, BRIGHT SUNLIGHT, HIGH CONTRAST, HIGH SATURATION, HIGH SATURATION, COLOUR, BLUE SKY BACKGROUND, DYNAMIC IMAGE, GREEN YOUTH, SUMMER ATMOSPHERE)\u3002<\/p>\n

The first one: she puts a pink skateboard vertically in front of her body, leans forward, hands on her side of the face compared to a pretty hand gesture, looks at the lens, and looks confident and dynamic\u3002<\/p>\n

The second image: she sat on her pink skateboard, leaning forward, holding a doll in her arms and looking to the camera with a close eye\u3002<\/p>\n

Third, she sits side by side on the ground, stretches out a hand as if she wanted to touch the lens, stretches out and looks nice and interactive\u3002<\/p>\n

Fourth image: She sits on a pink skateboard flat on the ground, with her hands on a lovely white doll near her cheeks, with her body on the right side, with her eyes open, and she shows a sweet, confident look\u3002<\/p>\n

\"4a4ed5c8j00t2ewsb00mcd000tf00dbm\"<\/p>\n

\"4ff3ed64j00t2ews6007id000u00140m\"<\/p>\n

\"a9f6bc72j00t2ews6005wd000u00140m\"<\/p>\n

\"21dce52aj00t2ews60059d000u00140m\"<\/p>\n

\"c27d4044j00t2ews60063d000u00140m\"<\/p>\n

2:Gothic combat maid<\/strong><\/p>\n

Consistency required in Figure 2: Young women, cold faces, sharp eyes, black and short shoulder hair and thin sea with a sense of disarray, dressed in a simple black modern combat style, with black short socks and thick-floored Martin boots, wearing a samurai knife, background roof roof roof, twilight time, strong contrasts, film-sension, dark-screening<\/strong><\/p>\n

Tips:<\/strong><\/p>\n

The generation of four group chart photo series requires consistency with the following elements\u3002<\/p>\n

Characteristics: Young Asian women with cold faces and sharp eyes. A black and short shoulder and thin sea with a sense of disarray. The make-up is clean and slightly offensive, emphasizing the sharpness of the eye\u3002<\/p>\n

Apparel: A simple black modern combat-style dress consisting of black sleeveless shirts and black tweeds with black socks and thick-down Martin boots. The whole style combines punk and campus elements\u3002<\/p>\n

Equipment: A very simple, completely black sword with a knife\u3002<\/p>\n

Background: The roof roof of the city skyscraper, at dusk. The sky is dark blue and the horizon is burning the sun-down orange cologne, in stark contrast to cold urban buildings\u3002<\/p>\n

Image aesthetics: Movie senses, sun-black photography, strong visual impact, iconic low-angle wide angle lenses, exaggerating visions, using twilight reverse light to create sharp edge light, high contrasts, cold tone, full of stories\u3002<\/p>\n

The first one<\/p>\n

Use very low view. She sat on a small concrete wall on the roof, leaning on her side, stretching one leg forward and bending the other. She stood behind her with one hand, with another holding the knife handles of the samurai, standing on her side, stretching out her body, with a cold eye and a provocative look, in the broad evening sky\u3002<\/p>\n

Second image (crawling):<\/p>\n

The camera is close to the ground, and it's seen with a very oppressive low angle. She was crouching, leaning forward and coming very seriously to the camera. She's got her hands on the blade, which is tilted in front of her, and she's pointed at the sky with her eyes locked in the lens as if she were looking at the target\u3002<\/p>\n

The third one<\/p>\n

Middle view, lower perspective. She stood on her back to the side of the lens, held her sword in her hands, placed her in the back of her waist, and made a classic guard posture. Her eyes crossed her shoulders, staring at a landmark skyscraper building far away, displaying the loneliness and vigilance of watchmen\u3002<\/p>\n

Fourth (silent):<\/p>\n

She sat on the floor of the roof, with her back against the concrete wall, and her legs folded at will. She holds her sword in her arms, like a partner. Instead of looking at the camera, she raised her head slightly, closed her eyes, as though she had passed her cheeks by the night wind on the roof, and the look had shed a rare amount of peace and tranquillity outside the cold\u3002<\/p>\n

\"c10589adj00t2ews6003bd000u00190m\"<\/p>\n

\"dbeb1355j00t2ews6005gd000u00190m\"<\/p>\n

\"8f1ee27fj00t2ews6003jd000u00190m\"<\/p>\n

\"4639b5fej00t2ews60056d000u00190m\"<\/p>\n

2. Graphic information + continuous drama generation<\/strong><\/p>\n

This example shows that behind this 4.0 model is probably the integration of large model optimization and the Vinnamon diagram, which shows a series of images that can be described as Agent automatic<\/p>\n

Image Information Generation<\/p>\n

This update has increased control over the Chinese text, and some simple hints can produce the text, and the layout is fine<\/p>\n

Phrasing (resolution 9:16):<\/p>\n

Handsome style, make a troupe of potatoes, troupes, troupes in Chinese, graphs, words<\/p>\n

\"f35535ej00t2ewsc00r2d000sw00evm\"<\/p>\n

\"2ff44628j00t2ewsp0238d000u001hm\"<\/p>\n

\"3d87cbbaj00t2ewsn01pud000u001hm\"<\/p>\n

Sequence scenario generation:<\/p>\n

the demands generated by this series are more of a test of the unity of character! before the banana model and dream 4.0, we had to use the confyui workflow or gpt4o to extend the consistency of characters or animated roles, which was rather difficult to draw\u3002<\/p>\n

And now 4.0 is able to produce a series of scenarios with a hint in which make-up, background, and character are largely consistent and rare<\/p>\n

Tips:<\/p>\n

Cute style, make a full set of graphics, take a picture + text description, and finish up to four\u3002<\/p>\n

the story is that the kids are separated from their parents in the woods, scared and helpless, and they're about to cry out. at that point, the good, sweet little creatures in the forest discovered ta-- the fireflies with a little light are like little lanterns. and so together, a hairy little squirrel came out with a nut in curiosity, and even a soft, shy plum deer came in\u3002<\/p>\n

They comfort and assist the children in their own unique ways, the fireflies gather into a shiny light belt, the squirrel jumps in the direction, and the swans fall down and invite the children to come. The children, led by the friends of the forest, passed through the bush and finally heard their mothers and fathers rushing and running into their arms. After a family reunion, the children looked back to the forest, and the little friends were shimmering in the shadows and waved their goodbyes\u3002<\/p>\n

\"1e83db9j00t2ews60062d000u001hm\"<\/p>\n

\"bf43e8cfj00t2ewsq01mgd000u001hcm\"<\/p>\n

\"976d4d6cj00t2ewsq01ved000u001hcm\"<\/p>\n

\"321a476dj00t2ewsp01kbd000u001hm\"<\/p>\n

3. Reality of image generation<\/strong><\/p>\n

didn't the gpt4o in the last few months have a vague self-portrait? the wind in the back is on the bean bag or dream 3.0 model\u3002<\/p>\n

Tips:<\/p>\n

Please draw an extremely common iPhone photo of yourself, with no clear subject or image, as if it were a snapshot. The photographs were slightly motion-altered, and the sunlight or shop lights were uneven, resulting in slight over-exposure. The angles are awkward and confusing, and the whole thing is a deliberate mediocrity, like a self-portrait that was accidentally shot when you picked up the phone in your pocket. The main event is _____, in the background\u3002<\/p>\n

This is dream 3.0 effect<\/p>\n

\"5e694528j00t2ewsj00r1d000tc0180m\"<\/p>\n

\"ee7f8b39j00t2ewsm0128d000q0018jm\"<\/p>\n

\"838f134ej00t2ewsk00qrd000u00140m\"<\/p>\n

\"74acd5b1j00t2ewsk00spd000tc0180m\"<\/p>\n

the following is the effect of today's version 4.0: although both models are very real, the version of 4.0 is even more amazing, and in version 3.0\/3.1 a more detailed description of \u201cextremely ordinary\u201d \u201cminor exposure\u201d is required in order to produce more closely to real-day images<\/p>\n

And 4.0 does not need to use additional instructions, or less, to create a true sense of quality (e.g. natural exposure, random images) of the daily scene by simply using a plain human like a hint (e.g., \"Students in high school take their own pictures\" and \"Women in cherry trees\")<\/p>\n

And there is a problem of \u201cfixed face model\u201d in versions 3.0 and 3.1, resulting in duplicate or similar stylized features of the person's face; 4.0 solves the problem, and the person's face is more diverse and further enhances the authenticity and uniqueness of the image\u3002<\/p>\n

Here are some of the illustrations:<\/p>\n

    \n
  1. an extremely ordinary iphone photograph of itself, without a clear sense of subject or image, is a random snapshot. the photos are slightly motion blurry, and the slight over-exposure caused by the uneven lighting of the classroom sun and the sun on the side of the window shows a deliberate sense of mediocrity, like a self-censorship that was accidentally taken while taking a cell phone from a school bag. it's about high school<\/li>\n
  2. Photos of yourself under white cherry trees, girls, looking at the lens, holding a transparent umbrella, bright background, daytime, long black hair, pink coats, an oblique image, a cherry tree for the future<\/li>\n
  3. an extremely uncharacteristic iphone photograph of itself, without a clear sense of subject or structure, is a snapshot taken by hand. a slight over-exposure, caused by a slight blurring of the photo, or by an imbalance in the sunlight or indoor light, presents a general sense of intentional mediocrity, as if it were a self-portrait accidentally taken from a cell phone in a pocket. the lead is coserrem\u3002<\/li>\n<\/ol>\n

    \"682bfdb1j00t2ewso0166ud000u00140m\"<\/p>\n

    \"f9326f2fj00t2ewsp01y6d000u00190m\"<\/p>\n

    \"440ec5a0j00t2ewsq01rd000u00190m\"<\/p>\n

    \"b7777ef1j00t2ewsq02jhd000u001hm\"<\/p>\n

    \"e763b636j00t2ewso01kd000u00140m\"<\/p>\n

    \"f403569aj00t2ewsp01wd000u00140m\"<\/p>\n

    Boys also come to produce a few results, and they can see that images are created in a series of different requests, and that the image of the three images that follow is similar, but not the same for each:<\/p>\n

    \"18df620ej00t2ewsb00h0d000tr00d0m\"<\/p>\n

    \"82fb8231j00t2ewsm0167d000u00140m\"<\/p>\n

    4. Accuracy of poster text generation<\/strong><\/p>\n

    The Chinese language of the posters is also good. Accuracy, aesthetics<\/p>\n

    For example:<\/p>\n

      \n
    • OSCAR-WINNING FILM POSTER DESIGN, RED IS THE MAIN COLOR, AND THE FESTIVITIES ARE HIGH. IT'S WRITTEN IN GOLD-COLOURED FONTS, \"THE QUEUE OF THE WORLD\" AND LABELED \"(1949-2025)\" TO WITNESS THE JOURNEY. IN THE MIDDLE OF THE ROLLS, THE LANDSCAPE OF THE STEREO-WATER ARCHITECTURE IS WELL PRESENTED, THE TRADITIONAL TOWERS AND KIOSKS ARE WRONG, THE TREES AND MOUNTAINS ARE PLUMBING AND THE DETAILS ARE RICH. THE RED SILK BELT FLOATS AND TWO GOLDEN BIRDS SOAR, GIVING THE IMAGE A SENSE OF MOTION AND LIFE. \"2025.10.1\" AT THE BOTTOM, COMBINED WITH THE WORDS \"CHANGE OF THE NATION\", \"SCRIPT OF THE GLORY OF THE TIME, DREAMS OF THE GREAT POWER,\" TO INTEGRATE THE TRADITIONAL ELEMENTS AND MODERN DESIGN, EASTERN AESTHETICS, EXTREME AESTHETICS, SUPER-SIMPLISTICS, FILM PIXELS, HIGH-LEVEL PIXELS, 32K, HDR<\/li>\n
    • China's country style is a large silhouette poster: the image is a resounding desert and ancient city wall, with a long-sleeved dancer in a gorgeous Han dress in the center of the picture, and a flying fairy in the back of a wall floats with the wind and silk as it appears. At the top there is the big words of the calligraphy style: \"The Millennium Sunshine, Dreams Back to the West,\" followed by the small words: \"The years are like sand, the art is like gold, the beauty of the wind of the Guardian.\" The whole color of the gold is warm and the silk and the tree of the Hu Yang echoes the silt image, the image is hierarchical and the image is extremely cultural\u3002<\/li>\n<\/ul>\n

      \"54f6d4bcj00t2ewsb00l0d000tm00cqm\"<\/p>\n

      \"00bd48a9j00t2ewsm01gd000u00140m\"<\/p>\n

      \"795<\/p>\n

      \"e0c82686j00t2ewso01v2d000u00140m\"<\/p>\n

      \"4bd29373j00t2ewso01k1d000u00140m\"<\/p>\n

      \"d0cee2a9j00t2ewsk00xvd000u00140m\"<\/p>\n

      Summarize<\/strong><\/p>\n

      Finish these three pieces, I'm straight to the guy\u3002<\/p>\n

      I USED TO PLAY WITH THE AI DRAWINGS, AND THE THING THAT HURTS THE MOST IS THAT I CAN'T GET THE PART. BUT THAT'S WHAT DREAM 4.0 IS ABOUT, AND IT REALLY GIVES ME HOPE FOR \"A-A-FACE,\" AND IT'S REALLY GOING TO MAKE COMICS AND MIRRORS\u3002<\/p>\n

      The trueness of the photographs and the accuracy of the posters are better than surprises\u3002<\/p>\n

      Overall:<\/p>\n

      The ability of the hints to follow better, to be consistent with character characteristics, to understand in depth is enhanced again, multi-graph output + group output, and Chinese cultural understanding and Chinese text generate a unique set\u3002<\/p>\n

      the disadvantage is that sometimes some of the hints are overwritten, there is a lack of beauty, and some of the 2k maps are not so clear as to be magnified, not so much as a true 2k map, and there are points that consume more, and one is a score, which was four\u3002<\/p>\n

      All right, that's it. Thanks for watching. I'll see you next time<\/p>","protected":false},"excerpt":{"rendered":"

      THIS IS TO MEASURE THE IMAGE GENERATION FUNCTION OF DREAM 4.0, IN FOUR PARTS: 1: GROUP IMAGE GENERATION, PRESERVATION OF HUMAN IMAGE CONSISTENCY 2: TEXT INFORMATION + CONTINUOUS DRAMA GENERATION 2: AUTHENTICITY OF IMAGE GENERATION 3: POSTER GENERATION AND ACCURACY OF TEXT SAVINGS IN SMALL ACTION: I FOUND THAT IF A DIFFERENT POSTER OR IMAGE, STYLE AND COLOR DISTRIBUTION, OR HUMAN FACE WERE GENERATED IN THE SAME BATCH, BECAUSE NOW 4.0 WOULD TAKE ONE SCORE TO PRODUCE A MAP (NOT KNOWN AS THE PUBLIC VERSION) AND FOUR POINTS TO GENERATE IMAGES DIRECTLY BY A HINT, WHICH IS QUICKLY CONSUMED, SO I ASKED AI TO PRODUCE FOUR DIFFERENT IMAGES IN THE HINT, OR JUST TWO IMAGES, SO IT WOULD BE A HINT = 1 MAP, AND IT WOULD CONTROL THE COST OF THE SCALE\/REDUCE THE DRAW, HIP-HIP, WHICH IS ALSO A MULTI-GRAM<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[149,144],"tags":[541,3149,2531],"collection":[],"class_list":{"0":"post-42812","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"hentry","6":"category-jiaocheng","7":"category-baike","8":"tag-ai","10":"tag-2531"},"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/42812","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=42812"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/42812\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=42812"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=42812"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=42812"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=42812"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}