Freddy wong

[Loriel.AI @ou_zhen599]

Land Rover

GPT- image 2 prompt（Rainforest）👇
Create an ultra-premium cinematic campaign image for the **Land Rover Defender**, designed as a **same-series top-down aerial action poster** with a stronger **rainforest mud-sprint** direction. The core image is a single Land Rover Defender blasting at high speed through a narrow rainforest trail, seen from a strict overhead view, while an enormous explosive trail of **mud spray, wet earth, mist, and debris** erupts behind it with extreme visual impact. The image must feel hyper-real, forceful, elegant, and editorial, like a world-class automotive campaign that transforms off-road motion into a sculptural environmental event.

Use a **strict top-down aerial composition** with a tall vertical frame. The Defender should sit slightly below center or in the lower-middle portion of the image, racing forward along a narrow muddy road carved through dense rainforest. The vehicle must remain clearly readable from above, with precise roofline, hood, glass, wheel arch geometry, and iconic Defender proportions. Behind the vehicle, build a **massive vertical mud-and-mist wake** that stretches far up the road and dominates the composition. This wake must feel like a violent ribbon of churned rainforest earth and wet spray, turning the trail into a monumental burst of motion.

The **mud plume** is the key visual hero:
near the rear tires it should be heavier, denser, darker, and more forceful,
then expand upward into rolling muddy vapor, wet dust, soil spray, shattered droplets, and earthy atmospheric bloom,
with a clear progression from thick ground-level mud impact to lighter mist and suspended particle haze.
The trail must feel large enough to create true poster-scale visual shock, almost like the car is detonating the rainforest floor behind it.

The Land Rover Defender must be rendered with exceptional product fidelity:
accurate proportions,
true-to-brand roof and body geometry,
premium paint finish,
realistic glass reflections,
authentic off-road tires,
precise trim, panels, and shadow shapes,
slight wet mud splatter naturally adhering to the body,
while still preserving a luxury automotive-advertising level of finish.
The vehicle must feel powerful, heavy, premium, and fully believable in motion.

The road and rainforest must be highly realistic:
a wet muddy trail with tire ruts,
uneven earth texture,
stones, roots, and broken soil,
dense rainforest vegetation on both sides,
lush dark green canopy,
bushes and foliage forming a narrow corridor around the road,
subtle mist and humidity in the air.
The forest should frame the road like a natural tunnel, amplifying the central motion line and making the Defender and mud explosion feel even more dramatic.

The interaction between **mud, water, mist, and debris** must be extremely sophisticated:
heavy mud chunks close to the tires,
wet soil spray expanding outward,
finer droplets suspended in the air,
light rainforest mist mixing into the upper plume,
soft atmospheric diffusion toward the far end of the trail.
Everything must feel physically correct, silky, and seamlessly layered.

Lighting should be soft, cinematic, and rainforest-natural:
filtered daylight passing through humid canopy,
soft highlights on the wet vehicle roof and body edges,
subtle reflective sheen on mud and water,
gentle volumetric light within the mud plume,
controlled shadow depth in the foliage,
smooth tonal transitions throughout.
The light must feel premium, natural, and refined, never harsh or synthetic.

Keep the branding layout minimal and premium:
**Land Rover** logo,
**DEFENDER** title,
and an elegant cinematic tagline such as:
“CONQUER THE WILD”
“POWER THROUGH THE STORM”
or
“BORN FOR THE UNTAMED”

Color palette:
deep rainforest green,
wet earth brown,
muddy umber,
mist gray,
soft gold daylight,
dark graphite vehicle accents.
The image should feel rich, natural, humid, powerful, and luxurious.

Overall mood:
top-down cinematic action, rainforest tunnel, high-speed mud sprint, explosive earth wake, premium off-road power, environmental force, same-series brand consistency, world-class automotive advertising.

Rendering style:
ultra realistic cinematic automotive photography, strict top-down aerial shot, Land Rover Defender racing through a rainforest mud trail, massive explosive mud and mist wake, hyper-detailed wet earth and debris simulation, soft filtered jungle light, premium automotive campaign, 8k, world-class advertising quality

Negative prompt:
static parked vehicle, weak mud trail, dry desert dust look, cartoon mud, cheap off-road poster, low-detail rainforest, plastic-looking car, cluttered composition, neon color grading, game UI aesthetic, generic SUV photo

[Sairah @Sairah_0]

GPT Image 2 On ChatGPT

Prompt:

Create an ultra-realistic editorial lifestyle photograph in a clean minimalist studio with a soft light gray seamless background.

The subject is a cheerful woman with a light blue pixie haircut. She is caught mid-run as if trying to escape, laughing uncontrollably with her mouth wide open, eyes happily squinting, and a playful, mischievous expression. Her body naturally leans forward in motion while being slightly pulled backward by the tension on her sweatshirt.

Pose :
Dynamic running pose.
One foot lifted naturally off the ground.
Arms swinging naturally while running.
One hand holding a colorful rainbow swirl lollipop.
Hair showing subtle movement from running.
Clothing and body posture should clearly convey forward momentum.

Outfit :
Oversized pastel blue sweatshirt.
Relaxed beige wide-leg trousers.
Red checkerboard Vans slip-on sneakers.
Round eyeglasses resting on top of her head only (not worn on the face).

Doodle Interaction :
On the left side of the frame, draw a simple black hand-drawn doodle character in a minimalist stick-figure style with:
rounded head,
tiny dot eyes,
straight expressionless mouth,
thick bold black marker lines.

The doodle stretches one long arm and firmly grabs the back of the woman's oversized sweatshirt, dramatically pulling the fabric backward to stop her from running away.

The pulling force should subtly shift the woman's upper body backward while she continues trying to run forward, creating believable physical interaction.

The doodle itself should appear to struggle slightly, with tiny motion lines around its hand and arm emphasizing the pulling force.

Add playful doodle motion lines around the woman's head to emphasize laughter and movement.

The doodle must appear as if it was hand-drawn directly over the finished photograph using thick black marker ink.

Hands :
Anatomically correct hands.
Natural finger proportions.
Sleeves remain oversized but both hands stay clearly visible.
Realistic grip on the lollipop.

Lighting & Style :
Bright high-key studio lighting.
Soft natural shadows.
Ultra-clean minimalist composition.
Editorial lifestyle photography.
Realistic fabric physics.
Humorous visual storytelling.
Premium modern aesthetic.
Playful, wholesome atmosphere.
Ultra-realistic.
Crisp details.
8K quality.
Vertical 4:5 composition.

[Freddy wong @FreddyWang98]

@Oo_Motoko 这是个写报告的 agent 吗

[Bin Liu @liu8in]

exactly why @HyperFrames_ is HTML-based

it's agents' native tongue, it's how LLMs express ideas visually

guess we aren't crazy and @claudeai team agrees

made this explainer using hyperframes, which is basically HTML + CSS + JS

coming to Claude MCP & Design soon!

[郭宇 guoyu.eth @turingou]

我有预感 2026 年会是我人生中极其重要的一年，在经历了二十多年孤单的自我叙述后，我对软件的理解与我对生活的表达因为 AI 与 vibe coding 开始走向交叉，融合，浸润。以前，我曾专注于做一些大家喜欢，老少皆宜的产品，从今年开始，我发现软件本身却成了自我实现的方式。我学习，成长，设定目标，试图在数十年的人生中不断挑战，活出新的自我，活好每一天，我从学习和成长中编写软件，并从语言的空间中理解各行各业真正的智慧与奥秘，我开始学习歌唱和跳舞，我也学习数学和如何独自一人于林间伐木。我慢慢意识到，从今年开始，仅依赖对话与叙述。我就可以成为任何人，只要我想。

[Unity for Games @unitygames]

Unity AI is now in Open Beta 🎉💫

We believe AI has the most impact when it helps creators move faster while staying in control of the creative process.

Use our built-in agent tuned for Unity workflows or connect the AI tools you prefer via AI Gateway and MCP Server.

[QuentinHsu @quentin_hsu]

一个原生 macOS 应用，用来集中收集、预览、更新并共享 AI Agent 技能，可用于 Codex、Claude Code、Cursor、GitHub Copilot 和自定义 Agent。

https://t.co/IdFbD6xk7v

Enjoy~

功能目前还很基础，欢迎反馈~

[Barret李靖 @Barret_China]

看到很多朋友问过一个问题，为什么给我的 Claude Code 安排任务，它都不会一口气执行完，而是跑最多几十分钟就停下来，然后问我要不要继续。例如让它把项目中的单测全部补全（大概 1k 个），它跑了大概 200 个就停下来了。

cc 并不是对一句话任务抗拒，如果不理解它的执行机制，很难设计出能跑长程任务的 harness 流程。

在执行一个超大任务的时候，单 agent 的执行流程大概是这样的：1）刚开始是高效模式，指令遵循效果特别棒；2）跑了大概 80k tokens 的时候，context 开始逼近 compact 阈值；3）紧接着，对话历史被压缩为摘要，模型开始忘记刚才修复单测的细节；4）再经过一两轮 auto-compact，它甚至会开始重复检查已修复的测试，当触发 maxTurns 并且 response 没有 ToolUse 指令时，模型会退出任务，然后开始询问用户："我已经修复了约 200 个测试，要继续吗？"

如果你在当前 session，回复继续，接下来的工作，它会做的更加不符合预期，并且退出得更快。

任何试图在一个 agent session 内完成海量工作的方案，最终都会碰到 context 膨胀 → compact → 信息丢失 → 效率下降的问题。

其实优化方向也特别简单，设计一个主-子 Agent 的运行模式（任务调度器），同时将任务进度写到 file system 中（进度持久化），每个子 agent 有独立 context、独立退出逻辑，主 agent 只负责调度和进度追踪，从而绕过单一 agent 的所有瓶颈。

因此给 cc 的指令需要包含至少这三部分：

1）任务分解。不要给一个无边界的指令（如修复所有单测），而是先扫描出所有失败测试，按目录或模块分组，每组 15-30 个，作为一个独立子任务。关键是每个子任务的 prompt 必须自包含——写清楚文件路径、错误现象、期望行为，不能写"根据之前的分析来修复"，因为子 agent 看不到父 agent 的历史。

2）进度持久化。在项目根目录维护一个 progress.json，记录 completed / failed / pending 三个列表。主 agent 每轮调度前读这个文件决定下一批任务，子 agent 完成后更新对应条目。这样即使主 agent 自己被 compact，重读文件就能恢复全部状态。

3）失败策略。子 agent 报错时，如果错误可修复，用 SendMessage 继续同一个子 agent（保留错误上下文更高效）；如果方向完全错了，启动新的子 agent 避免锚定在错误路径上；多次失败则上报用户，不要无限重试烧 token。

Claude Code 其实已经内建了这套能力。最直接的方式是启用 Coordinator Mode（输入 /coordinator），主 agent 自动变成纯调度者：它不执行任何实际工具调用，只负责理解子 agent 的返回结果、合成下一步的具体指令、并行派发独立任务；而每个子 agent 会通过 AgentTool 启动，它们有独立 context。

记住一句话就行了：设计多个 agents，各司其职、快进快出，把进度交给文件系统来记忆。

[sarah guo @saranormous]

Caught up with @karpathy for a new @NoPriorsPod: on the phase shift in engineering, AI psychosis, claws, AutoResearch, the opportunity for a SETI-at-Home like movement in AI, the model landscape, and second order effects

02:55 - What Capability Limits Remain?
06:15 - What Mastery of Coding Agents Looks Like
11:16 - Second Order Effects of Coding Agents
15:51 - Why AutoResearch
22:45 - Relevant Skills in the AI Era
28:25 - Model Speciation
32:30 - Collaboration Surfaces for Humans and AI
37:28 - Analysis of Jobs Market Data
48:25 - Open vs. Closed Source Models
53:51 - Autonomous Robotics and Atoms
1:00:59 - MicroGPT and Agentic Education
1:05:40 - End Thoughts

[Freddy wong @FreddyWang98]

@SamBCui 大佬，怎么拿

[Freddy wong @FreddyWang98]

@binghe @0xAA_Science 全退款了吗

[Ring Hyacinth @ring_hyacinth]

项目开源啦！

▶ 完整项目：https://t.co/Bqf8JtbiyB
▶ Skill：https://t.co/SLToARPhVk

项目简介：
1. OpenClaw 龙虾的“像素办公室”：龙虾会根据状态自动走到不同位置（休息区 / 工作区 / bug 区）
2. 左下角添加他昨天的工作小记。
3. 支持邀请其他龙虾加入办公室（丰富功能开发中）
4. 手机端适配

created by: @simonxxoo and me

[Freddy wong @FreddyWang98]

@cccchuizi 牛逼

[玩个锤子 @cccchuizi]

【开工季福利来袭！！！】
为了感谢大家对我们的支持，我们打算免费送一台 Mac Mini256GB + 2000刀PackyAPI额度，助力2026年VibeCoding 和养殖龙虾！

一等奖：macmini 256GB 一台
二等奖：Packyapi额度100 刀✖️10 人
三等奖：Packyapi额度50 刀✖️20 人

点赞+评论+转发即可参与抽奖！(中奖后我们会仔细检查哦)

3 月 8 号截止～

网站链接：https://t.co/82YMcSw72H
相关文档：https://t.co/nqVvEmipwU

[尤雨溪 @yuxiyou]

AI 让 10x eng 真正变成 100x eng

[Freddy wong @FreddyWang98]

@songguoxiansen 蹲一下

[Freddy wong @FreddyWang98]

@94vanAI @grok @grok hey，反推一下视频提示词，我想复刻这个视频

[Freddy wong @FreddyWang98]

@Deeman_R @iamfkmpro @0xROAS

[Freddy wong @FreddyWang98]

@Deeman_R @iamfkmpro @0xROAS contact me

[Freddy wong @FreddyWang98]

@li12826221 期待分享