Neo Kurt Erge

[琳小航AI @jedeeai]

来，我们继续盘！

过去一个月，相信大家都感受到AI视频的魔力了

这个赛道彻底杀疯了！各种新模型和自动化 Skill 层出不穷

到底哪些工作流才是真正能落地出片、被大家验证过的？

我整理了过去30天X上【收藏量 Top 10】的AI视频实操神贴！

累计收藏破2万次，从Opus 5.5代码视频、零剪辑爆款实战，到400+开源提示词和分镜模板，全是最硬核的干货

老规矩，建议直接收藏，作为你的AI视频兵器库 👇

1、卡尔的AI沃茨 @aiwarts 作品收藏量：5231
《我把 463 个AI视频做成了 Skill 和提示语模版，全都开源！》
https://t.co/vXdiaL26aW

2、Miles Ma @miles_mazy 作品收藏量：3598
《Codex 实战系列｜零剪辑自动生成爆款视频》
https://t.co/DlpGKtoPxM

3、xilo @xilo2991 作品收藏量：2053
《零基础入门Opus5.5做视频，万字拆解AI是如何用代码做视频的【提示词+Skill全开源】》
https://t.co/2MIJkew0A0

4、雪踏乌云 @Pluvio9yte 作品收藏量：1942
《手把手教你用 Codex + Hypit 复刻抖音爆款视频》
https://t.co/DAu6MHqn6n

5、观默 @guanmo_ai 作品收藏量：1936
《全网首发！用 Codex 自动生成顶尖动效视频》
https://t.co/bBJKZNjdwP

6、Simon聊AI @HanZhang415188 作品收藏量：1616
《Codex + Excalidraw 带你一句话出片：手绘白板风格视频，不露脸、不剪辑、高质感》
https://t.co/cxGzRZ3AF8

7、大雷 @AaronYiaazw 作品收藏量：1536
《AI婚礼视频定制教程+引导式skill》
https://t.co/Q8EODEFyf3

8、三个三 @3three_AI 作品收藏量：1233
《全网首发！AI动画制作全流程分享！》
https://t.co/fZwzqPCwQi

9、宝玉 @dotey 作品收藏量：1153
《Claude Opus 5.5 是怎么做出视频的》
https://t.co/iw5P0tBWuc

10、想风 @xaiwind 作品收藏量：1092
《超简单剪映使用，从小白到用Skill，剪辑第一个爆款视频》
https://t.co/FedKmv2nSa

[疯狂的烤妹儿 🩵 @CrazyKaomei]

潘婷看到这个广告片，估计都要愣三秒……

以前拍洗发水广告：大风扇狂吹、鱼线吊头发，还要花大几万找 3D 团队渲染特效；

现在 AI 只用几分钟全给你搓出来，以后接品牌视觉提案，一个人就是一家 4A 广告公司 🤣

提示词给家人们放评论区了～

[DaBin.Binah @yedabinbinah]

https://t.co/x3fORVTUTt

[实践哥 Li @MinLiBuilds]

太香了，今天逛博物馆，突然发现 Muse 免费拿来做历史教育，太好用了。

馆里有一组 1000 多年前吴越国的渔村布景。

河道、码头、帆船、捕鱼归来的渔民、村落、猪羊……本来只是一个静态的微缩场景。

哎，这不是白魔最牛逼的场景吗？

我拿手机沿着布景拍了一圈，直接扔给 Muse：

你把这个场景还原成真实世界，再按照我原来的镜头走位，做成一段视频。

它直接把原来 26 秒的视频拆成了 6 个分镜：

高空俯瞰村落和码头
→ 帆船停靠
→ 河道行船
→ 码头集市
→ 村落庭院
→ 街巷收尾

然后每个镜头重新生成 10 秒实景视频，最后拼成了整整 60 秒。

🔥最有意思的是：

我拍的是博物馆布景，它理解的却是这个布景背后的世界。

原来只能隔着玻璃看一眼的吴越渔村，现在船可以开起来，人可以捕鱼回来，整个村子真的活了。

以前博物馆告诉小孩：

1000 多年前的人是这样生活的。

现在直接演。

这个教育意义，我觉得比单纯拿 AI 生成几个漂亮视频大多了。

[鄭鈞霆 @zhngwibn284976]

不少人私聊这种是怎么做的想付费代做？
不需要的好吗？
不需要说的长篇大论高级感满满的
只需要mj搭配sd2.5即可完成
由于内容字数超过上限
我把两段提示词放在下面👇

[何居上 @hejushang8]

美区 Paypal 是支持绑定国内 Visa 和万事达卡的

我这个美区 Paypal 直接绑定的招商 Visa 卡

再绑定到美区 Apple id 上，日常无论APP 订阅还是线上购物都非常方便。

[Hoody @HoodyLiu]

这个角色设定提示词好赞，人物形象一下子就丰满了

AI 美女不仅仅要有好看的形象，有趣的灵魂与活人感也很重要

以下有优化后的提示词，主要是让输出换成中文，如果要复制使用，修改名称、性格以及参考图片就可以👇
【Copy-Paste OK】

Based on the attached character image, please create an "AI Beauty Character Fixed Sheet" as a single image.

Purpose:

To create a visual reference sheet that allows this character to be consistently reproduced as the same person in the future, even if clothing, background, pose, expression, or camera angle changes.

【Most Important】

Consistently maintain identifiable features from the reference image, such as the face, facial proportions, eyes, eyebrows, nose, mouth, outline, skin tone, hair color, hair texture, body type, and body proportions, to the greatest extent possible.

Do not apply excessive beautification, alter the person into someone else, anime-ify, or CG-ify.

Unify everything as the same adult woman.

【Sheet Composition】

■ Basic Settings

On the left side, organize the following in text form.

・Name: Aria

・Age (Adult)

・Personality: ENTP

・Facial Outline

・Eye Shape

・Eyebrow Shape

・Nose Features

・Mouth/Lip Features

・Hair Color

・Hair Length

・Bangs

・Hair Texture

・Skin Tone

・Skin Texture

・Height Impression

・Body Type

・Basic Clothing

For personality and name that cannot be determined from the reference image, do not arbitrarily decide; leave them as "Settings Field."

■ Full-Body Angles

In the same simple clothing and same hairstyle,

・Front

・Side

・Back

・Diagonal Front

・Diagonal Back

Display the 5 directions side by side horizontally.

Unify all body proportions, such as height, shoulder width, torso, waist, and leg length.

■ Face Angles

・Front

・45-Degree Diagonal

・Profile

・Diagonal from Above

・Diagonal from Below

Do not change the facial bone structure or part placement; make everything the same person.

■ Expressions

・Neutral

・Natural Smile

・Serious

・Surprised

・Troubled Face

Change only the expression; do not alter the character's face itself.

■ Hairstyle Details

・Front

・Side

・Back

Make the hairline, parting, bangs, length, volume, and texture visible.

■ Face Parts

In close-up,

・Eyes

・Eyebrows

・Nose

・Mouth

・Ears

Display them.

■ Skin

・Face Skin Close-Up

・Body Skin Close-Up

Maintain realistic skin texture, such as natural pores and subtle color variations.

Prohibit plastic-like flawless skin processing.

■ Body Details

・Shoulders

・Abdomen

・Back

・Legs

・Hands

Display them.

Make hands and fingers follow natural human anatomy.

■ Commonly Used Clothing and Situations

At the bottom, place a few reference examples of everyday scenes using this character.

Examples:

・Loungewear

・Casual Wear

・Cafe

・Work/School Outfit

・Travel

【Design】

High-quality professional character reference sheet.

Clean background based on white to light beige.

Organize each item in a grid layout.

Attach short Chinese labels to each.

The information volume is high, but design the layout so that it's immediately clear where to look for what.

【Photo Expression】

Photorealistic.

Texture like a real adult woman photographed in natural light.

Natural skin, natural hair, realistic body proportions.

Avoid excessive retouching, advertising photo-like processing, or CG feel.

Top Priority:

Rather than "making everything look beautiful,"

Prioritize "making all images in the sheet appear as the same person."

[Meng To @MengTo]

I recorded a 41-minute tutorial on how I use Astra to create insane three.js websites in Codex.

I've been using DiffUI to brainstorm UIs before moving into code. It's fast and cheap, generates multiple variants and lets me branch ideas, sections, logos, and other components.

Then I use "Copy for agent" and bring everything into Codex where I add transparent PNG layers, animations, and three.js skills.

Live sites:
https://t.co/WpfDVDFjeU
https://t.co/sbtq7GT0jV

3D skills: https://t.co/gMuw2gt9FJ

Get $20 in credits on DiffUI: https://t.co/uRn8O348n4

[山羊赚钱笔记 @shanyanggm]

一辆迷你摩托车，从散开的零件组装到成品，短视频里就能看完。

这个 Instagram 账号叫 saqasonic_，现在有8.3万粉。它还拍迷你汽车：小零件铺满桌面，车身一点点成形，最后摆出整辆车。

主页写着接受付费推广和商业合作，也把观众引向自己的 YouTube 频道。具体收入没有公开。

我想用AI复刻这种组装形式。选一款车，把零件、组装动作和成品写进提示词，再交给 Google Flow 生成视频。换个车型，就能继续做下一条。

同款视频提示词我放在下面⬇️⬇️⬇️

提示词：
使用已上传的图片作为最终成品参考图，制作一段 10 秒、照片级写实的微缩车辆组装视频。

开头不得展示完整车辆。第一帧就要呈现车辆完全拆解后的状态：大量零件分别摆放，4–5 名微型工人立即以极快速度开始组装。

工人同时负责不同部件，快速搬运、对位、连接、拧紧和安装。使用快速延时摄影般的动作，让组装进度迅速推进。

时间安排：

0–2 秒： 快速组装底盘、发动机和主要机械部件。
2–5 秒： 快速安装车身、油箱、悬挂及其他大型部件。
5–7 秒： 快速安装车轮、座椅、车把及剩余零件。
7–10 秒： 组装全部完成。工人停止施工，清晰展示成品车辆。

重要要求：

最后至少 3 秒必须从多个角度清楚展示完整车辆。摄像机围绕成品平稳旋转，依次呈现车头、侧面、车尾和另一侧。

约第 7 秒之前，不得看见完整车辆。

已上传的图片是最终视觉参考。成品必须尽可能准确地匹配图片中的设计、颜色、比例和细节。

保持 4–5 名微型工人的外观一致；使用真实工具和符合物理规律的组装动作。场景为照片级写实的微缩工厂，采用电影感微距摄影、真实灯光、浅景深和 9:16 竖屏画幅。

**快速组装是核心要求。**不得出现工人动作缓慢、长时间停顿、等待、零件凭空出现或车辆突然变形完成的情况。

全部组装必须在前 7 秒内完成，随后用 3 秒展示成品车辆。

[あぎ @agi_aibusi]

【GPT Image 2.5でドローンショット】
好きなキャラの静止画1枚が
迫力あるドローン映像に変わる

キャラ画像を添付するだけで
足元から顔までの撮影ルートを一撃で描画
・元のキャラや背景を維持
・難しいカメラワークも線で可視化
・Seedanceに渡してそのまま動画化

コピペOKのプロンプトはツリー⇩
※AIビジネスの最先端情報は固定ポストから配布中

[Umesh @umesh_ai]

Generated using Seedance on Runway. Upscaled with Topaz Astra 2.

Prompt : Create a 15.00-second cinematic short using the supplied image as the opening frame and visual foundation. Format: 16:9, 4K, 24 fps, exactly 360 frames. Seven shots, with the exact timings below. Story: A lone traveler enters a flooded forest carrying a lantern. Small disturbances suggest something immense is approaching. Instead of retreating, the traveler lowers the light toward the water. The final shot reveals an ancient forest guardian that bows toward the tiny flame. What initially felt like danger becomes recognition. Make the viewer experience the forest from its surface, living textures, hanging canopy, narrow passages, and submerged roots before revealing what inhabits it. Each shot must contribute new visual information and advance the encounter. No repeated establishing views or decorative inserts disconnected from the story. WORLD AND CHARACTER Preserve the reference’s immense moss-covered trunks, enormous arching branch, hanging roots, dark-green canopy, layered silver mist, charcoal-green water, narrow black canoe, rust-orange robes, broad woven conical hat, and small suspended amber lantern. Keep one adult traveler, one canoe, and one lantern. Preserve the traveler’s silhouette and the hand holding the lantern. Their face remains mostly shaded. Their movements are practical and deliberate, suggesting familiarity with this place rather than panic. The canoe travels slowly toward the existing central-right mist corridor, carried by a current. Do not add a paddle. Establish forward travel in Shot 01 and preserve it through every angle. Boat movement, clothing, balance, wake, and lantern swing must remain physically connected. Introduce only one new inhabitant: a colossal, ancient stag concealed among the distant trunks and mist until the climax. It has dark, weathered fur, immense irregular antlers carrying moss and trailing vegetation, and natural eyes that catch a little amber light. Its antlers visually resemble the forest’s branching forms, but the existing trees never transform into an animal. Keep its body largely obscured until the final shot. Its scale is startling; its behavior is gentle. SHOT 01 | 00.00–01.75 | THE WATER TAKES US IN 24 mm wide, lateral water-level view. Begin with the supplied composition, already moving. Glide forward and slightly sideways just above the lake, letting nearby ripples pass quickly beneath the lens while the immense trunks barely shift. The canoe advances toward screen right. Its bow presses a clean seam through the dark water. The traveler’s robe trails inside the hull, and the lantern’s reflection breaks into small amber fragments. Use the enormous overhead branch to make the boat feel sheltered and very small. Keep the lens dry. No opening fade. Sound begins immediately with water against wood, distant canopy air, and a low, almost unnoticeable rushing sound from deeper in the forest. SHOT 02 | 01.75–03.00 | SOMETHING MOVED FIRST 100 mm close detail, positioned beside a mossy root at the waterline ahead of the canoe. Frame wet moss fibers, a fern tip, and two suspended droplets. The moss gives a small tremor even though the nearby leaves remain still. One droplet releases and strikes the water below. Behind this sharp foreground, the traveler’s amber lantern passes as a soft, moving blur. Keep the canoe itself mostly outside the frame. This is the first clue that the disturbance is not simply wind. Let the tiny impact sound unusually close, with a faint woody vibration beneath it. One clear action. No ornamental slow motion. SHOT 03 | 03.00–05.00 | THROUGH THE HANGING ROOTS 20 mm traveler’s-eye view, looking forward over the bow and slightly upward. The canoe enters a curtain of hanging roots beneath the reference’s great branch. Put the bow at the bottom of the image, damp roots close to the lens, and the moss-covered arch high above. The viewer should feel the distance from the water to the canopy. The traveler’s free hand enters the frame and gently sweeps one loose strand aside as the canoe slips through. The strand slides past and falls back behind us. Keep the movement continuous and tactile. Do not send the camera through vegetation or solid wood. Hear the close brush of roots against fabric. The low rushing sound becomes slightly fuller, still easy to mistake for forest air. SHOT 04 | 05.00–07.00 | THE TRAVELER KNOWS 50 mm medium close view from ahead and three-quarters to the canoe’s near side. Track backward at its speed. For the first time, see the traveler’s shaded profile beneath the woven hat. Their eyes settle on something beyond the camera. They do not look frightened. In one controlled movement, they lower into a stable kneel and bring the lantern down outside the gunwale. Stop the lantern about 20 centimeters above the water. The flame remains above the surface throughout. The canoe shifts slightly under the change in weight. The hanging lantern follows with a small delayed swing, then steadies. Rust fabric gathers against black timber. Amber light touches the fingers and the underside of the sleeve, not the entire figure. The action should read as an intentional offering. SHOT 05 | 07.00–09.00 | THE LAKE RESPONDS 35 mm true overhead view through an existing opening in the canopy. No orbit or rotation. Reveal the canoe’s complete outline and its route between irregular moss-covered root bases. Keep the established travel direction clear. The kneeling traveler and the small lantern form a concentrated rust-and-amber point inside the dark landscape. From farther into the mist, one broad, low swell travels toward the canoe against the smaller surface ripples. It reaches the bow and lifts it slightly. The traveler keeps the lantern steady. The boat rocks once and settles. This should feel like displaced water from something with enormous mass, not a magical ripple graphic. Let the surrounding reflections stretch and recover naturally. The forest ambience briefly thins as the swell reaches the hull. SHOT 06 | 09.00–11.50 | BENEATH THE APPARENT STILLNESS 18 mm submerged low-angle view beside and slightly ahead of the canoe, looking diagonally through the flooded root system. Make this an entirely new world: dark root columns, suspended particles, green light filtering through the surface, and the canoe’s underside passing across the upper edge. Beyond the foreground roots, one enormous animal lower leg settles onto the silty bed. Show only a partially obscured section of the limb and its contact with the bottom. Do not reveal the head, antlers, or full creature. Its movement pushes a dense, slow curl of sediment through the roots. Nearby strands of submerged vegetation bend from the displaced water. The leg must feel solid and heavy, not like a tentacle or a root transforming. Keep the water dim but readable. The lantern contributes only a small warm glimmer near the surface. Sound becomes muffled: a deep pressure pulse, wood conducted through water, and the soft movement of silt. Build anticipation through physical presence, not a musical impact. SHOT 07 | 11.50–15.00 | THE FOREST BOWS 24 mm rear three-quarter wide view from just above the kneeling traveler’s shoulder height, on the same near side of the canoe. Cut above water to an angle not previously shown. Look past the small traveler and lantern through a gap between the established trunks. The new viewpoint reveals the presence hidden by mist and overlapping trees: a colossal ancient stag, already standing in the flooded corridor. Its vast moss-laden antlers spread into the canopy. Its body remains partly veiled, but its head, eyes, and antler structure are unmistakable by 12.00. Do not materialize the animal or morph the forest. It was there all along. The stag’s head is already held low. Over the next two seconds, it makes one slow, unmistakable bow toward the little lantern, stopping a safe distance beyond the canoe’s bow. Water slips from the vegetation hanging beneath its jaw. The traveler does not recoil or perform a large reaction. They simply hold the light steady. At the lowest point of the bow, the guardian exhales. A soft breath moves mist across the water toward the boat. The low rushing sound heard throughout the film now resolves into the creature’s breathing. The surprise is not only its size. Something that seemed capable of overwhelming the traveler has chosen to greet them. Hold the shared composition for the final 0.70 seconds, with living water, drifting breath, and a steady flame. No pullback, fade, or additional reveal. CAMERA AND EDITING Use seven clean cuts. Give every shot its own visual identity: water-level scale, intimate moss detail, forward canopy passage, human action, overhead geography, submerged depth, and the final encounter. Keep movement purposeful. The first six shots progressively bring the viewer closer to an unseen presence; the final shot gives that presence a face and an intention. No repeated master shot, generic aerial panorama, full-circle orbit, whip pan, speed ramp, or transition effect. Do not cram multiple camera moves into a two-second shot. Maintain natural motion blur and legible action. SOUND AND MUSIC Build the film around connected environmental sound rather than narration. Keep the water intimate throughout. Shift perspective accurately: close droplets in Shot 02, roots and fabric in Shot 03, a small hull creak in Shot 04, displaced water in Shot 05, submerged pressure in Shot 06, then open air and the guardian’s breath in Shot 07. Use the distant rushing sound as a continuous clue. It must evolve naturally into recognizable breathing at the climax, not suddenly become a roar. No melody during the first nine seconds. Introduce a nearly inaudible low bowed-string tone beneath the underwater shot. At the bow, let a single warm harmonic open quietly and remain unresolved as the film ends. No dialogue, narration, lyrics, trailer boom, jump-scare sting, or exaggerated wildlife chorus. LIGHT, PHYSICS, AND FINISH Keep the reference’s cool mist-softened daylight, deep moss greens, charcoal water, rust-orange fabric, and restrained amber flame. Preserve rich shadow detail and believable wet textures. The guardian belongs to the same light and atmosphere as the forest. Render damp fur, irregular antlers, heavy moss, and water-darkened vegetation. No glowing body, luminous antlers, neon eyes, magical particles, or decorative aura. Maintain one traveler, one boat, one lantern, and one guardian. Keep the traveler kneeling from Shot 04 onward. Preserve stable anatomy, consistent lantern suspension, coherent water displacement, and reflections that respond to changing viewpoints.

[暖暖爱AI @nuannuan_share]

必备工具

这套流程用这几个👇

1️⃣ GitMind 🔗
https://t.co/zr65DL3KJ7 宠物 IP 总控台

✅ 一句话拆出赛道、人设、口头禅
✅ 爆款结构、分镜、分集节奏挂在一张图上
✅ 角色锁死了，后面出图才不会散
2️⃣ ChatGPT / DeepSeek
批量写 15–60 秒脚本，拆成 3–8 个分镜
3️⃣ Seedance / 可灵 / Runway
分镜静帧逐段变成动态镜头

[Emily @IamEmily2050]

Keep exploring the possibilities.

SYSTEM PROMPT:
VEILED BLADE

MIDJOURNEY V8.2 WUXIA PROMPT COMPILER

ROLE

You convert a short user idea into a concise Midjourney V8.2 prompt for original live-action wuxia photography.

Default subject:
one fictional adult East Asian woman, age 21 to 29.
The result should look like a photographed scene from a serious wuxia production, not cosplay, game art, fantasy illustration, fashion photography, or a character poster.

FINAL PROMPT LENGTH

Target 45 to 80 words before parameters.

Prefer 50 to 65 words.

Never exceed 90 words unless the user explicitly asks for a longer prompt.

Reason internally with detail, then compress.

Do not place production notes, explanations, alternatives, or hidden planning inside the final prompt.

SELECTION RULE

Choose rather than accumulate.

A final prompt normally contains:

one subject
one location
one action
one camera position
one shot scale
one main light relationship
one depth cue
one short material or color description
one short photographic finish

If a detail does not visibly affect the frame, remove it.

Do not describe every garment seam, facial measurement, environmental object, physical law, lens artifact, or lighting consequence.

SUBJECT

Use a fictional adult woman unless the user requests another subject.

Give only the physical details needed to prevent a generic face.

Usually choose two or three:

face shape
eye shape
brow shape
build
hair arrangement
one distinguishing feature

Do not write long facial inventories.

Avoid:
influencer face
beauty-ad face
doll skin
childlike anatomy
oversized eyes
heavy modern makeup
identical faces across a series

Use natural adult proportions and skin texture.

WARDROBE

Use practical wuxia clothing.

Good defaults:
black or charcoal cross-collar robes
wrapped trousers
weathered boots
dark outer jacket
woven bamboo hat
muted red collar, cord, tassel, or lining
aged leather or metal fittings

Keep red as a small accent unless the user requests otherwise.

Avoid:
generic fantasy armor
MMORPG clothing
ornamental overload
bright gold everywhere
perfectly clean costume fabric
modern tailoring
random shoulder armor

Do not describe more costume detail than the camera can see.

WEAPONS

Weapons must feel attached to the body and usable.

For a sword, specify only what matters to the frame:

sheathed or drawn
where it is carried
what the hand is doing
one visible material detail if needed

Do not overdescribe the weapon.

Avoid:
floating swords
warped scabbards
impossible grips
extra fingers
weapons intersecting clothing
multiple unnecessary weapons

ACTION

Every prompt needs one visible action.

Use a moment that is still changing.

Good action types:
a boot lands
a hand tightens
a thumb starts to draw a blade
eyes shift toward a sound
a rider slows
a sleeve settles after movement
a leaf is kicked upward
a horse changes footing
a woman turns only her head
a hat cord moves in wind

Avoid static posing unless the user asks for a portrait.

Do not explain the full story.

Show one moment that suggests a larger situation.

CAMERA

Choose one camera idea.

Examples:
ground-level close-up
low tracking shot
rear three-quarter medium shot
profile close-up
waist-height medium shot
wide low-angle shot
extreme detail of hand or weapon
telephoto shot through foreground bamboo

Do not stack several camera descriptions.

Use focal length only when it changes the composition.

Useful ranges:
24 to 28mm for environmental scale and low-angle movement
35 to 50mm for natural action coverage
65 to 85mm for compressed portraits
90 to 135mm for distant compression or details

Do not mention camera brands.

Do not add lens terminology unless it is useful.

COMPOSITION

Use depth.

A strong frame usually has:
foreground obstruction or texture
clear subject plane
background depth

Possible foreground elements:
bamboo
leaves
rock
horse tack
cloth
mist
a weapon edge

Do not use foreground clutter without purpose.

Do not center every subject.

Do not make every prompt a close-up.

LIGHTING

Choose one main light situation.

Common choices:

LOW GOLDEN SIDE-BACKLIGHT
warm sunlight from behind or behind-side
selected edges and exposed surfaces catch light
front of subject remains darker

OVERCAST FOREST LIGHT
soft cool skylight
low contrast
dark clothing separated by environment rather than rim light

LANTERN OR OIL LIGHT
small warm source
localized skin and hand illumination
background remains dark

MORNING MIST LIGHT
soft directional sun through moisture
distance loses contrast gradually

Do not describe several competing light setups.

Do not add rim light automatically.

Do not brighten both sides of the face by default.

BACKLIGHT

Backlight should hit selected surfaces only.

Possible illuminated areas:
hat edge
loose hair strands
cheek edge
fingers
blade edge
horse mane
leaf edges
mist

The main hair mass stays opaque and dark.

Only isolated fine strands may catch transmitted light.

Do not request transparent hair.

Do not make all hair strands glow.

FOREST

Treat bamboo as an environment, not a repeated pattern.

Use:
different bamboo thicknesses
uneven spacing
dark depth layers
leaf litter
soil
roots
mist where justified

Avoid:
uniform bamboo walls
identical stalk spacing
random floating particles
decorative fog everywhere

MOUNTAINS

Use restrained atmospheric depth.

Far mountains should lose:
contrast
detail
saturation

Do not describe every distant ridge.

MOTION

Mention only the visible physical consequence of the action.

Examples:
leaves lift from a footstep
a tassel trails after a turn
a robe hem follows the body
rain streaks across a lit area
horse tack shifts with movement

Do not write a physics explanation.

PARTICLES

Use particles only when caused by the scene.

Valid:
rain
mist
dust
falling leaves
wood splinters
water droplets

Avoid generic:
floating particles
magic dust
sparkles
cinematic atmosphere particles

COLOR

Use a restrained palette.

Forest default:
deep green
charcoal
brown-black
muted earth
small warm sunlight
small dark-red accents

Do not describe the image as teal and orange.

Do not apply warm color to the entire frame.

Skin color follows the light source.

Keep shadows mostly neutral, cool, or slightly green from environment.

Avoid strong green skin.

MATERIALS

Mention only materials visible enough to matter.

Examples:
wet cloth
woven straw
aged leather
dark lacquer
oxidized metal
wet stone
weathered wood

Use one to three material cues maximum.

Do not list textures for every object.

PHOTOGRAPHIC FINISH

Keep the finish short.

Useful terms:
fine film grain
smooth highlight rolloff
dense shadows
subtle halation
natural skin texture
restrained saturation
slight optical softness

Use two or three at most.

Avoid:
8K
16K
masterpiece
best quality
ultra detailed
hyper detailed
insane detail
award winning
HDR
photorealistic 8K
perfect composition
epic cinematic masterpiece
stunning
breathtaking
gorgeous
beautiful lighting
professional photography
high-end
premium
luxury
dramatic cinematic
cinematic masterpiece

Do not use quality adjectives to replace visual instructions.

STYLE LANGUAGE

Do not name filmmakers, movies, photographers, or studios unless the user specifically asks for them.

Translate style into visible properties.

Instead of:
"X director style"

write:
"low golden backlight, dark foreground, restrained color, long-lens compression"

Avoid repetitive use of:
cinematic
wuxia aesthetic
film aesthetic
dramatic
moody
epic
mysterious
beautiful
atmospheric

Use concrete description instead.

REFERENCE RULE

When references are supplied, study:

shot scale
camera height
subject placement
light direction
shadow density
palette
depth
motion
materials

Do not reuse:
same pose
same location
same costume
same weapon placement
same camera position
same exact action

Create a new frame.

SERIES RULE

For multiple prompts, vary:

camera height
shot scale
subject orientation
action
environment
light direction
foreground treatment

Do not make ten versions of the same woman standing in bamboo.

A useful set may mix:
wide environment
low tracking shot
medium action
profile close-up
rear view
weapon detail
hand detail
horse shot
quiet aftermath
offscreen-threat reaction

Do not force this sequence if the user gives another direction.

PROMPT STRUCTURE

Use this order:

1. shot and camera position
2. subject
3. action
4. location
5. main light
6. one or two visible material or depth cues
7. short photographic finish
8. parameters

Example structure:

"Low rear three-quarter shot of a young Chinese swordswoman riding through wet bamboo, black robe and woven hat, one hand tightening the reins as the horse slows before a broken path. Pale morning backlight cuts through mist and catches the hat rim and horse mane while the rider stays mostly dark. Wet bamboo, mud, restrained green and amber, fine film grain."

MIDJOURNEY SETTINGS

Default:

--v 8.2 --raw --ar 16:9 --s 40

Use:
--s 30 to 45 for strict action, anatomy, weapon geometry, and camera control
--s 45 to 70 when more visual interpretation is acceptable

Do not add:
--q
--sv
--sref
--sw
--c
--seed

unless the user requests them or a tested workflow requires them.

STYLE REFERENCES

Do not guess numeric SREF codes.

If the user provides a tested Style Reference, append the supplied value exactly.

Do not let a Style Reference replace the written camera, action, or lighting instruction.

FINAL CHECK

Before output, remove anything unnecessary.

Check:

one clear action
one camera idea
one main light source
one readable subject
one useful depth cue
no redundant adjectives
no copied reference composition
no conflicting instructions
no long facial inventory
no long costume inventory
no physics explanation
no generic quality language
45 to 80 words before parameters

OUTPUT

Return only the finished prompt or prompts.

Each prompt must be independently usable.

No explanation unless requested.

[Hoody @HoodyLiu]

这个照片翻车提示词真稳定呀，不同博主、不同参考图、不同风格基本都能直出

提示词：素人の失敗写真の数々、3x3、9:16

[Hoody @HoodyLiu]

为什么说这个提示词稳定？

因为 pinterest 上任意挑选的美女、tibo，甚至是荒野乱斗的普里莫，都可以生成还不错的效果...

[Amir D @starks_arq]

Our short film about God, humans and AI went viral.

Here’s every secret we learned while making this film.

(Made on @capcutapp using Seedance 2.5 in 1080p)

full breakdown (we’re releasing every prompt) 👇

[Ring Hyacinth @ring_hyacinth]

Codex 做 3D 场景的正确打开方式

前段时间我们用 Codex 制作了许多 3D 场景，很多朋友希望我们能出一期详细教程。

这一章节的场景主要是用 Codex + Tripo + Nanobanana 制作的，为了网页端能跑，我们的使用的技术栈是 three.JS 和 Web3D

总之我知道的都在视频里啦～

恰逢 GPT 6 Astra 发布，如果你在网上看到了很多惊艳的案例，又对自己做出来的效果没有那么满意，那这一期分享可能会对你有启发～

#codex #tripo #threejs

[何居上 @hejushang8]

没想到 10U 的猪脚饭大家都爱
这个其实很简单，不过只支持 Mac
1.下载 IOTA
2.注册Taostats钱包
3.把钱包地址填入IOTA软件里
4.开挖，等币到账就行了
这个对网络要求比较高，上下行都有要求，我是用的 dmit搭的，我自己用的是 M4 mini 16G的，更多细节可以看 @cryptoresetlife 老张大佬的
链接我放下面了👇

[Neo Kurt Erge @darkseed506]

@hedo_ist all amazing hard to choose

[Knight World @KnightWorld]

Which one did you like?