MJ.

AI Art

[MJ. @MadnessJ_Yan]

Whose cat decided the moon was a toy?! 🌙🐈

Clouds need brushing, the moon needs polishing, and fallen stars won’t pick themselves up.

Why is there more housework in the sky than on Earth? ☁️✨

#ai #Dreamcore #AIGC #AIVideo #DigitalArt #aiart

[0xBiggoose @0xbiggoose]

这是哪个币圈人才搞的，这不比网剧还精彩

[Beginnersblog @beginnersblog1]

MiniMax H3 is so cheap

It is priced at roughly one-third of Seedance 2.0 and is also more affordable than premium models such as Seedance 2.5. That means creators can test more ideas, generate more versions, and improve their videos without burning through the entire budget.

Since the model is open-weight, it can also be deployed locally. This gives creators and small teams more control over their workflow and makes ongoing experimentation much more practical.

[Umesh @umesh_ai]

Another educational animation prompt with MiniMax H3!

Prompt : Create a 15-second cute, funny, educational creative animation video for kids ages 3 to 7. Concept: Teach the letter B using the word “Butterfly” with two major visual highlights: 1. a slow, satisfying, highly engaging transformation from the word “Butterfly” into a cute butterfly 2. a slow, magical transformation from the stylized animated world into a fully realistic natural scene with a realistic butterfly The video should feel beautiful, colorful, charming, educational, and visually mesmerizing. The transformations must not feel rushed. They should be the centerpiece of the video and should unfold clearly enough for children to enjoy and understand. Main learning message: “B is for Butterfly.” Visual style at the beginning: Use a premium kid-friendly animated style with soft pastel skies, rounded shapes, bright happy colors, smooth motion, clean composition, playful details, expressive design, and a light sense of humor. The early part should feel like a beautiful storybook mixed with polished modern animation. Animation priorities: The word-to-animal transformation must be gradual, elegant, readable, and creatively staged. Children should be able to see how the letters slowly become parts of the butterfly. The stylized-to-realistic transformation must also happen gradually and beautifully, as if the animated world is blooming into real life. Timing and scene flow: 0:00–0:02 Open on a soft dreamy pastel background with floating flowers, tiny sparkles, and gentle clouds. A large uppercase “B” appears in the center with a smaller lowercase “b” beside it. Both letters bounce softly into place with a cheerful and friendly feeling. The narrator says slowly and clearly: “B is for Butterfly.” Hold the letters long enough for recognition. 0:02–0:04 The word “BUTTERFLY” writes itself slowly across the screen in large rounded colorful letters. Each letter fills with a different beautiful pattern, such as gradients, dots, stripes, floral shapes, and soft wing-like textures. Keep the word fully readable and centered. Let the audience admire the full word before the transformation begins. 0:04–0:09 This is the main highlight. The word “BUTTERFLY” transforms slowly and satisfyingly into the butterfly. The transformation must be gradual, staged, and visually delicious to watch: The “B” gently opens outward and unfolds into the top wings. The “U” stretches and curves into one lower wing. The first “T” and second “T” slim down and merge into the butterfly���s long body. The “E” softens and reshapes into decorative wing markings. The “R” curls upward into one antenna. The “F” reshapes into the second antenna with a playful wobble. The “L” and “Y” arc outward and become the final wing edges and tail details. Important: Do not transform everything at once. Let each part morph one after another in a smooth chain reaction. The letters should bend, curl, unfurl, rotate, stretch, and softly glide into their new forms. The child should still sense the original word during the transformation. Use sparkling trails, flowing color transfer, and soft elastic motion to make it deeply satisfying. The morph should feel magical, clear, and creatively intelligent. 0:09–0:11 The butterfly is now a fully formed cute stylized cartoon butterfly. It has lovely colorful wings, a tiny smiling face, big friendly eyes, and soft funny charm. It flutters gently in a playful little loop, hovers above a flower, almost sneezes from pollen in a cute funny moment, then recovers and smiles. Show the word “Butterfly” below it briefly for reinforcement. The narrator repeats: “Butterfly.” 0:11–0:14 This is the second major highlight. The entire animated scene slowly transforms into a realistic world. The butterfly flaps its wings in slow motion. With each flap, realism increases gradually: its wing textures become more detailed the colors gain natural variation the body becomes more lifelike the flowers gain realistic petals and depth the background softens into a real sunlit garden the lighting becomes natural and golden the air gains subtle movement and realistic depth of field This transformation should feel breathtaking, seamless, and immersive. Do not cut abruptly from cartoon to realistic. Instead, let the animated world gently bloom into reality in layers. The viewer should feel they are witnessing imagination turn into nature. 0:14–0:15 End on a beautiful realistic butterfly resting on a real flower in warm sunlight. Display the text: “B is for Butterfly” The narrator says warmly: “B is for Butterfly.” The butterfly gently flutters once, then lifts off lightly. Style and quality requirements: Keep the letter B and the word “BUTTERFLY” clearly readable before and during transformation. Make the word-to-animal morph slow, rich, and visually rewarding. Make the stylized-to-realistic transformation equally slow, smooth, and magical. Use polished motion design, soft easing, beautiful color transitions, gentle humor, and strong visual clarity. Maintain a child-friendly tone throughout. Use clean typography, stable spelling, and seamless morphing. Audio direction: Use cheerful, soft, magical music. Add delicate sparkles, soft pops, flutter sounds, a tiny comedic sniffle or sneeze, and a gentle emotional swell during the realism transition. Voiceover should be warm, clear, slow, and easy for children to follow. Avoid: rushed transformations, unreadable letters, clutter, hard cuts, scary expressions, excessive camera motion, chaotic effects, or anything visually confusing for children.

[Framer 🇱🇹 @Framer_X]

Seedance 2.5 has serious competition.

Meet Hailuo MiniMax H3.

Head-to-head test below 👇

Which one wins?

[Gadgetify @Gdgtify]

An old Midjourney prompt of mine upgraded for GPT. I gave it a fruity subject

Prompt:

2x2 grid, 16:9 do this for 4 fruit made up creatures and 4 challenging poses, Generate [SUBJECT] as a humanlike pose figure performing [POSE_SYSTEM]. Rules: - infer SUBJECT_DNA automatically - construct the body entirely from subject-derived materials, parts, textures, or related components - do not simply draw a human wearing the subject; the subject itself becomes the body - infer BODY_MAPPING logically: head from a compact recognizable component torso from the strongest central form arms and legs from elongated or flexible elements joints from small round, hinge-like, seed-like, bead-like, or modular parts hair/accessories from strands, leaves, fibers, peels, cables, threads, sparks, or inferred details - create POSE_COUNT panels showing different poses - each pose must be anatomically readable and physically balanced - preserve the subject’s recognizable texture and material identity - use a clean lifestyle environment with soft light, simple background, and subtle shadows - no hard-coded fruit, pose, body part, accessory, room, plant, color, or prop unless supplied

[AIZC @blade2019runner]

天香宝境0021号
月色压低云城，万盏灯从桥上升起，像有人替长夜点亮归途。她们穿过柳岸、雨幕与群鹤翻飞的清晨，也曾并肩坐在日蚀之下，看云海收拢成巨大的漩涡。白衣倚着青衣，酒盏在掌心微暖；红伞划过风雨，衣袖被长风牵向远方。星轨转动，瀑布坠入无底深谷，宫阙在晨昏之间一次次醒来。可她们最珍惜的，并不是天象与奇观，而是有人在灯火将尽时仍靠在肩头，有人在万里星河倾覆之前，轻声说：今晚不必回去。
made by
@midjourney
@dreamina_ai
#AIart️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️️ #seedance #fantasy #erotic

[Zephyra Leigh @ZephyraLeigh]

seedance 2.0 in 4k is absolutely mind-blowing. 🤯🔥

[澜森 @lansenai]

非原创，Seedance2.5 30s直出现在这种风格也可以完美拿捏了。

[NΞXUS @NEXUS_TO_NOVA]

Last time, I showed how one failed Seedance 2.0 generation burned through my credits and how expensive Seedance 2.5 could become for everyday AI creators. This time, I flipped the script. Instead of using a normal prompt, I gave the model a full shooting script with timed beats and strict constraints. What you’re about to see is three raw generations placed back-to-back, with no editing within the clips. Each one is a single continuous shot.

[Smiling Khan @AIwithkhan]

UGC on Seedance 2.5 is on another level

Awesome Results

Prompt 👇

Use the uploaded reference image as the exact character reference. Preserve her facial identity, hairstyle, eye color, makeup, skin tone, body proportions, white sleeveless fitted top, light blue wide-leg jeans, pearl choker, rings, and bracelets consistently throughout the video. Use the uploaded sunglasses, retail box, and leather carrying case as locked product references. Maintain perfect product consistency, including the frame shape, lenses, hinges, colors, materials, and proportions.

Create an ultra-realistic UGC luxury creator review filmed inside a modern luxury bedroom with warm golden-hour sunlight, soft natural shadows, and a premium lifestyle aesthetic. The camera feels like a handheld smartphone with subtle natural movement while maintaining cinematic commercial quality.

The video begins with the woman sitting on the bed beside the retail box and leather case. Smiling at the camera, she says, "I genuinely wasn't expecting to love these this much." She picks up the box, opens it naturally, reveals the leather case, then slowly removes the sunglasses while continuing, "The packaging already feels incredibly premium."

She rotates the sunglasses slowly in front of the camera, showing the frame, hinges, and lenses as natural reflections glide across the surface. She smiles and says, "The finish feels amazing, and they're incredibly lightweight."

She puts on the sunglasses, stands up, and walks toward a large full-length mirror. Looking at her reflection, she adjusts the frame naturally and says, "Honestly... they look so good, and they're really comfortable on the eyes, even in bright sunlight."

She turns slightly left and right so the sunglasses catch the sunlight from different angles before removing them with a smile. Walking back to the bed, she places the sunglasses beside the leather case and retail box, then picks them up one last time and holds them beside her face.

Looking directly into the camera, she smiles warmly and says, "Definitely one of my favorite accessories this year." The camera slowly pushes in on the sunglasses before fading out.

Ultra-realistic UGC fashion content, authentic creator review, cinematic handheld smartphone movement, luxury bedroom, macro product cinematography, realistic reflections, detailed frame textures, expressive facial animation, perfect lip sync, shallow depth of field, premium color grading, 4K HDR, 16:9, no subtitles, no logos, no watermarks, no on-screen text.

[虎小象 @hx831126]

最近很火的天宫视频，我研究出来一套矩阵式提示词。我的思路如下：

1我没有具象化天宫要怎么画，而是利用模型的随机性，

然后给模型一个总意境

天上之宫，不落人间；
人如一笔，宫阙无涯；
近处有巨物，远处有天境；
静处显庄严，动处生仙气；
空而不寂，满而不塞。

目的是丢给图像模型后，一次性输出 9 张不一样的图片，基本是次次都不一样，张张都不一样。

输入提示词给图像模型（我这里应用的是 GPT），它就能自动输出 9 张图片。次次不一样，张张不一样，人人不一样，

在创作提示词的时候，其实经历了好多好��轮呢，几个调整之后的关键点就是天宫的。地面不要有护栏，地面要有镜面感，人物和建筑的���例要让看过去有巨物感，天空要挂着月球或者银河之类的东西，总之第一次你拿去自己可以去更多的优化

提示词👇

[Noura @noorwithwifi]

Made with Seedance 2.0

Prompt:

A cozy Studio Ghibli-inspired cinematic cooking animation showing a delicious grilled chicken shawarma wrap and refreshing mint-lime soda being prepared step by step. Begin with slicing juicy shawarma meat on a vertical rotisserie, cutting fresh limes, adding ice, mint, and sparkling soda to a chilled glass, then assembling a warm tortilla with grilled chicken, lettuce, tomatoes, onions, purple cabbage, and creamy garlic sauce before rolling it into a perfect wrap. Finish with garnishing the drink with mint and lime, showcasing fizzy bubbles and the final plated meal on a rustic wooden counter in warm golden lighting, with highly detailed food, smooth camera movements, soft depth of field, and a magical hand-painted Ghibli aesthetic.

[Hoody @HoodyLiu]

最近的中国天宫系列 AI 视频非常火，基本都是mj垫图+seedance/kling生成。

但不是所有天宫作品都具备高级感，我发现跑出圈的这些不只是因为场景宏大，还有很多构图/色彩技巧。

构图上，用树木、岩壁和廊柱形成天然画框；用极小的人物对比巨型建筑与云海；中轴、对角线和S形动线交替使用，让每个镜头既稳定又有纵深。

配色则非常克制：大面积低饱和蓝灰、象牙白打底，只用少量朱红和橙金点亮视觉中心。远景继续降饱和，前景加深对比，空间感自然就出来了。

[INK @0xInk_]

some new fighting tests completely out of context

and yes it's still better on Seedance 2.0 rather than Seedance 2.5

[Emily @IamEmily2050]

Seedance V2.5

I love drama more than action clips because you can tell more stories and explore people emotion.

I will update the Skill and add Seedance V2.5 when I feel I got hold of all the tricks.

Japanese live action TV drama style, photoreal. A young couple in their mid 20s explodes into a vicious fight in their cramped Tokyo apartment at night. Acting-first: raw emotion, believable blocking, micro expressions. Simple locked camera.

[CHARACTERS] YUI (female lead): 25-year-old East Asian woman, tired modern-city face; warm natural skin tone with real pores, faint dark circles; reddened teary eyes, furrowed brows, nostrils flaring with fast breathing, bitten lower lip, jaw tight; eyes full of resentment, fury and despair, locked on him; dark-brown mid-length hair in a messy claw clip, loose sweaty strands stuck to her cheeks and neck; oversized worn beige knit cardigan over a faded white cotton tee, real home wrinkles and slight pilling; thin frame, shoulders trembling with anger and crying, the aura of someone finally exploding after long suppression.

KENTA (male lead): 27-year-old East Asian man, unkempt gamer face; pale indoor skin tone with real pores and light stubble on the jaw; narrow unfocused eyes, puffy lids, black blue-light glasses, mouth slightly open, expressionless; eyes cold, dismissive, looking at her like a stranger, impatient and completely unable to empathize; greasy messy black hair, fringe falling over his eyes; baggy pilled dark-grey hoodie with a loose stretched collar and a faint stain; slouched, sunk deep into the sofa, the aura of selfish obliviousness.

[SETTING] Night, cramped Tokyo one-room apartment, cluttered: unwashed dishes and empty cans on the low table, delivery boxes and laundry piled in the background. Dim light: one warm floor lamp near the sofa vs cold blue TV glow.

[TIMELINE dialogue must be spoken in pure Japanese, no English audio] 0s-5s: Medium locked shot. Yui enters carrying heavy plastic shopping bags, stops dead seeing Kenta sprawled gaming; her chest heaves. Kenta's eyes glued to the screen, fingers mashing the controller.

YUI (exploding, sharp screaming voice): 「ちょっとケンタ！またゲーム？！なんでいつも私ばかり？！」 KENTA (without looking, utterly dismissive): 「んー、今いいとこ。静かにして。」

6s-15s: Close shot, slight handheld drift. Yui steps in front of the sofa, blocking the TV with her body; both hands clutch the plastic bags, knuckles white; tears burst out. Kenta annoyed, tilts his head to see around her, shrinks deeper into the sofa, eyes still trying to reach the screen. YUI (screaming through sobs, whole body shaking): 「ご飯も買い物も掃除も全部私！あんたはゲームばっかり！ちょっとは手伝おうと思わないわけ？！ねえ？！なんで私ばっかり？！」 KENTA (interrupted mid-game, impatient shrug): 「だってさ、お前が楽しそうにやってたじゃん。やる気出なかったし。」

16s-25s: Close-up. Yui's face shifts from fury to disbelieving heartbreak; her grip gives out, the bags drop to the floor with a dull thud; she covers her mouth, tears flood, shoulders convulse. Kenta stays slouched, only a slight frown. YUI (fully broken, hoarse, sobbing uncontrollably): 「楽しくなんかない！毎日毎日、もうつらいよ！なんでわかってくんないの！」 KENTA (cold, staring at her like she's a nuisance): 「…は？泣いてんの？めんどくさ。」

26s-30s: Wide locked shot. Yui turns and walks to the kitchen, thin trembling back. Kenta finally pauses the game but does NOT get up; he scratches his greasy hair, yawns. The physical and emotional distance between them maxed out; hold.

[GLOBAL] No subtitles, no BGM, only live location sound: ragged breathing, cracking sobs, cloth rustle, bag thud, controller clicks. Film-grade realistic lighting, warm vs cold contrast. Real natural micro-expressions, no exaggerated oily AI acting, no face morphing, no extra fingers, no random on-screen text.

[BMX @bmx_ai13]

Wait for the turn... the sleeves understood the assignment. Which moment wins, the spin or the final close up?

Seedacne 2.0 Prompt: Create a photorealistic 15 second cinematic fashion video in 16:9, filmed as one continuous natural take at an elegant tropical resort beside a turquoise swimming pool at dusk. A stylish young woman stands beneath hanging banyan roots and dense glossy green foliage. Warm amber lights from a distant outdoor lounge glow softly behind her. A large handcrafted ceramic urn sits near the walkway. The setting feels humid, lush, intimate, and genuinely lived in.

She wears a fitted blue and ivory paisley tunic with a square neckline, a small tied keyhole at the chest, sculpted waist seams, and dramatic flared trumpet sleeves. Pair it with loose ivory trousers, delicate gold drop earrings, and dark hair arranged in a neat low bun secured with a small gold claw clip. Her makeup is refined and natural, with warm skin, subtle eyeliner, and softly glossed lips.

Use a stable tripod camera at upper torso height with a 45 millimeter full frame lens, moderate depth of field, realistic skin texture, accurate fabric detail, and gentle handheld micro vibration only. Keep the woman centered while the wide frame reveals the pool, tropical plants, tiled path, and warm background lights. Lighting should combine soft warm frontal illumination on her face with cool cyan reflections from the pool and deep natural green shadows. Avoid beauty filter plasticity. Preserve believable pores, fine hair, sleeve weight, cloth folds, and consistent anatomy.

From 0 to 3 seconds, begin in a medium close shot as she maintains direct eye contact and takes two relaxed steps backward, gradually revealing the full tunic and trousers. Her hands settle lightly near her hips and the sleeves sway from real momentum.

From 3 to 5 seconds, she shifts her weight onto one hip, gives a subtle head tilt, and begins a smooth turn toward camera right. Her expression remains poised and self assured.

From 5 to 7 seconds, she completes the turn to briefly show the back of the outfit and the gold hair clip, then rotates naturally through profile and returns to face the camera. Her earrings catch small warm highlights and the flared cuffs trail slightly behind each movement.

From 7 to 10 seconds, she performs a restrained rhythmic sleeve flourish. Both forearms rise to chest level, cross lightly, open outward, and circle down with soft wrists. The gesture feels spontaneous and fashionable, never like a rigid dance.

From 10 to 13 seconds, she steps toward the camera while extending one hand close to the lens, creating natural perspective and shallow foreground blur. Her face transitions into a flattering close shot. She glances upward for a moment, then returns her eyes to the viewer.

From 13 to 15 seconds, she leans slightly toward the lens and finishes with a warm confident smile, holding the final expression long enough to feel intentional. End on the live image with no platform end card.

Use physically accurate motion, consistent facial features, stable garment patterns, correct fingers, natural blinking, subtle breathing, and realistic hair movement. No cuts, no slow motion, no text, no logos, no watermark, no visual glitches, no warped hands, and no artificial camera orbit.

Add an original upbeat fashion pop instrumental at approximately 129 BPM with a clean kick, soft bass, light hand percussion, airy synth textures, and subtle wordless vocal accents. Synchronize the turn, sleeve flourish, forward step, and final smile to clear musical beats. Include faint natural poolside ambience beneath the music.

[古一 @MANISH1027512]

同样的提示词，这是gpt的结果。

这是一个从gpt-image2从上线就被问了无数次的问题：

为什么这类提示词 GPT-Image 2 生成的图片噪点很重很脏？

这张图正好就是一个很典型的案例，通过这个案例我一次给你讲清楚，为什么gpt不行而midjourney/即梦却可以。

1.这类提示词gpt-image2为什么不行？

问题不完全在于提示词太长，而在于高频细节指令堆得太密。

提示词里同时出现了：

超写实、精细三维渲染、高清纹理、HDR、16K、UE5、Octane、粒子发光、烟雾弥散、体积光、电影级质��……

这些词都在要求模型不断增加纹理、锐度、反光、裂纹、火星和粒子。

当画面又包含大量黑色暗部、蓝色火焰、橙色裂纹、烟雾和碎石时，模型会用大量细碎纹理填充暗部，最后形成典型的 detail overload：

黑色肌肉变成岩石裂纹
烟雾变成密集颗粒
火焰边缘产生大量碎亮点
画面看起来又锐、又糊、又脏

尤其是“16K、超高清、HDR”这类词，并不会真正提高输出分辨率，反而可能让模型在有限像素中硬塞更多伪细节。

想减少噪点，最有效的方法，是先找到决定画面的核心关键词锚点，再配合少量辅助描写。

以这张图为例，真正影响画面的核心锚点其实只有：

暗黑超写实、极低机位、钟馗、巨型地狱战马、幽蓝火焰、电影级体积光。

剩下的内容只需要补充主体体型、动作、配色和环境关系。

当 UE5、Octane、HDR、16K、高清纹理、粒子、烟雾、锐化等同类词不断叠加，模型会持续增加表面纹理、局部反差和碎片细节，画面自然容易变脏。

先用少数关键词锚点确定视觉母体，再用少量描写控制主体与构图，通常比堆满“画质增强词”更加稳定。

2.为什么Midjourney/即梦可以？

核心原因，是两者处理提示词的方式不同。

Midjourney 更擅长识别一组关键词共同指向的视觉母体。

比如：

超写实、UE5、Octane、HDR、16K、体积光、高清纹理、电影级质感、粒子特效……

这些词看起来很多，但它们大多指向同一个成熟的审美方向：

高规格暗黑奇幻 CG、游戏概念艺术、史诗级电影画面。

Midjourney 通常会把这些同方向的词进行合并和压缩，将它们理解成一组整体风格信号，而不会给每个词单独寻找一个视觉落点。

所以最后呈现出来的效果往往是：

材质语言被统一
光线方向被统一
色调和对比度被统一
相似的风格词被自动合并
冲突较弱的词被重新平衡
不重要的细节可能直接被舍弃

它更像一位审美主导的视觉导演。提示词提供方向，模型负责取舍，并优先保证整张图在视觉上成立。

这也是为什么 Midjourney 很“吃风格词”，也就是我们说的“叠词”。

所以 Midjourney 看起来像是“每个词都吃到了”，实际上很多词只是共同把画面推向了一个它非常熟悉的风格区域。

[Kiki @Mayz1169]

just having fun with Seedance 2.5

prompt in thread 👇

[Kōda @aimikoda]

Tried fitting 50 cuts into 30 seconds with Seedance 2.5.

I let codex count the cuts. Including flashes, explosion/splash effects and the final fade-out as transitions, the video has 42 cuts/transitions in total. Technically, it contains 36 actual editorial cuts. You can check the prompt below.

Created with Seedance 2.5 on @astorie_ai