Pinned Tweet
2006 年開始拍人像,出過三本《制服.女孩》。現在用 AI 做寫真集,一本 36 張,每天在這裡發一張。追蹤我,一張都不會漏。
Portraits since 2006, three Uniform Girl books. Now 36-frame photobooks made with AI, one frame a day. Follow so you don't miss one.
1
1
216
Swanky Hsiao | swanbear.eth retweeted
A brief history of AI Coding since Nov 2025:
Nov 2025: Opus 4.5 releases, OpenClaw gets built
Dec 2025: community agrees that models get good enough to handle 95% of coding tasks
Jan 2026: Ralph loops get popular, early efforts on loops/orchestration
Feb: Everyone feels knackered, too much strategic decision-making, fears around cognitive debt begin
Mar: Skills explosion as community converges around easy-to-share processes (superpowers already ahead of the game here)
Apr: most SOTA devs haven't written a line of code in months
May: "The unreasonable effectiveness of HTML" - more interest in richer review surfaces
Jun: "Loop Engineering" wins vaguepost of the year, signals more hype around process
Jul: Every talk in every conference becomes "here is my software factory"
Aug: Opus 5 speaks incomprehensibly, interest in ASD-STE100 peaks
Sep: Grok Bot gets popular, more interest in long-running "manager" threads which retain context and delegate work
Oct: The promised speed gains haven't materialised, "dark factory" feels a long way off still
Nov: No-slop manifesto releases, reading the code still appears essential
TL;DR: no-one is writing code anymore, and we are still figuring out what that means
拍這張,我會靠近眼神,把紙海的弧線留在周圍。你第一眼看見她,還是那片海?
I'd frame her gaze closely, keeping the paper waves around her. What catches your eye first: her or the sea?
1
2
36
《把一頁書,讀成一片海》完整 36 張在這裡:
One Page Becomes a Sea, all 36 images:
swanky.github.io/photography…
13
「AI 平均臉」不只出在五官,也出在相機。
我把自己一本寫真集攤開數過:主角鏡 32 張,平視 28 張,站著沒動作 16 張。臉每張都對,翻起來還是像同一張。
改五官提示詞之前,先數數你那組有幾張是平視。
The "AI average face" isn't only in the features. It's in the camera too.
I counted one of my photobooks: 32 character shots, 28 at eye level, 16 just standing there. Every face was right, and the set still looked like one picture.
Before rewriting the face prompt, count how many of yours are at eye level.

ALT 第一版四格圖。史旺姬穿水手服,分別在霓虹街捧南瓜燈、糖果店拿棒棒糖、屋頂月光下坐著、派對中比手勢。多為相近鏡位的半身或近景;右下有背景人形。 First-version grid: Swanky Ji in a sailor outfit holds a pumpkin lantern on a neon street, holds a lollipop in a candy shop, sits on a rooftop, and poses at a party. Most frames use similar camera heights and medium or close framing. Background figures appear at bottom right.
纯文生有一个缺点就是,人像很容易趋于平庸——五官精致但毫无辨识度,就是大家常说的“AI 网红平均脸”。
想要打破这种精致的塑料感,其实不用死磕五官提示词,把重心放到“环境光影 + 人物动态”上:
1.用动态破除僵硬:比如图里“一手扶草帽、衣袖随风鼓起、侧头远望”的瞬间抓拍感,比正脸直视镜头更有故事。
2.用光线代替磨皮:给足逆光、林间丁达尔光、发丝发光等光影条件,皮肤质感自然就出来了。
3.给一点点垫图/参考权重:哪怕只是借一点真实摄影的构图和色调,生出来的氛围感都会直接拉开一个维度。
1
29
我怎麼把「姿勢」拆成九個欄位來檢查:
How I break "pose" into nine fields to check:
teapawt.net/swanky/status/21164945…
12
AI 說「我測過了」,那只是一句話。
Lauren Tan 的團隊改看操作影片、截圖和狀態圖。我把她投影片上的 42 種測試驗證逐項做成卡片:是什麼、以前誰負責、AI 之後怎麼做。
swanky.github.io/technical/a…
42
my AIE talk is up! come hear how the grok @bot team ships so quickly without compromise on quality
youtu.be/O_zcc6Xl12g
Swanky Hsiao | swanbear.eth retweeted
Announcing Artificial Analysis Coding Agent Index v2.1, upgrading FrontierCode and Terminal-Bench. Claude Fable 5.2 and Claude Opus 5.5 share the lead at 58
The Coding Agent Index measures models in agent harnesses; v2.1 of the Index upgrades FrontierCode from 1.1 to 2.0 and Terminal-Bench from 4.0 to 4.1.
Upgraded:
➤ FrontierCode-AA 2.0 (replacing FrontierCode v1.1): our implementation of Cognition's FrontierCode 2.0, comprising 132 real-world coding tasks graded on whether a repository maintainer would merge the agent's change
➤ Terminal-Bench 4.1 (replacing Terminal-Bench 4.0): the same 66 tasks, scored with updated verifiers that improve reliability and resistance to reward hacking
Staying in the Index: SWE-Atlas-QnA-AA
Key results:
➤ Claude Fable 5.2 and Claude Opus 5.5 share the lead at 58: Claude Fable 5.2 scores 58 at both xhigh and high in Claude Code. Claude Opus 5.5 (max) scores 58 in Claude Code and, paired with GPT-6 Luna High, in Cognition's multi-model harness Devin Fusion CLI. Claude Fable 5.2 (max) scores 56 as more of its attempts hit task time limits. Claude Sonnet 5.5 (max, 56) in Claude Code and Gemini 4 Carbon (high, 55) in Antigravity CLI follow
➤ Index scores are lower in v2.1, driven by the upgrade to FrontierCode-AA 2.0: models score lower by 5 points on average. Codex with GPT-6.1 Terra (max) is the only configuration which didn’t fall between versions, sitting within 1 point of its prior score
➤ Cost per task varies ~30x among configurations scoring 48+: from $1.51 for Codex with GPT-6.1 Sol (max, 52) to $44.22 for Claude Code with Claude Fable 5.2 (max, 56)
I'm wearing CLONE X - X TAKASHI MURAKAMI as my PFP on OpenSea opensea.io/collection/clonex…
Swanky Hsiao | swanbear.eth retweeted
oncyber never went dark. now it’s on forever.
i first fell for bitcoin back in 2014. the idea of a freer internet, where people could own things and have more control over their digital lives, stuck with me. years later NFTs brought that idea together with creativity, and i fell in love all over again.
at the same time, it was COVID. we were all far apart, and i wanted a place where i could hang out and have fun with my friends, wherever we were.
oncyber came from both of those desires: a free canvas for creativity, where people could connect and know that what they made was truly theirs.
what happened next exceeded my wildest dreams. 800,000+ creators, tens of millions of visitors… but what made oncyber special wasn’t the numbers. it was the people: people from all over the world who made spaces, shared their creativity, met each other and had fun together.
along the way, we raised venture capital and got caught up in the ambition of building a billion-dollar company. i started measuring my own worth against it, and for a while i lost touch with myself. tough way to live. then the market changed and the company had to wind down. it was hard but the hard moments brought me back to who i am (and for that, all praise be to God).
myself, the team and our investors couldn’t accept that what people had created should disappear because the company did. so for the past two years i’ve kept it running, and those worlds stayed alive.
now i’ve rebuilt it w all my heart and opened up the code, so anyone can build on it and take it further. the place people pour their creativity into doesn’t have to depend on anyone.
i don’t have a business goal for this. i just want to give people an open place to create, connect and have fun. this past month has been the most fun i’ve had building since the early days of oncyber, and i can’t wait for you to feel it too.
oncyber is forever.
Swanky Hsiao | swanbear.eth retweeted
reminder to run `/doctor prompt-audit` in your Claude Code repos if you haven't already :)
第一章 裂土裡的一點綠。一圈乾土,中間只有一小片濕泥。她蹲下來數種子。《稻穗成熟以前》,第 15 冊。
Ch. 1, A Green Point in Cracked Earth. A ring of dry soil, one patch of wet mud. She kneels to count seeds. Before the Rice Ripens, vol. 15.
1
33
























