惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Martin Fowler
Martin Fowler
V
Visual Studio Blog
有赞技术团队
有赞技术团队
T
Tailwind CSS Blog
B
Blog
I
InfoQ
博客园 - 三生石上(FineUI控件)
阮一峰的网络日志
阮一峰的网络日志
F
Fortinet All Blogs
H
Help Net Security
博客园 - Franky
宝玉的分享
宝玉的分享
博客园 - 司徒正美
C
Check Point Blog
G
Google Developers Blog
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
Jina AI
Jina AI
T
The Blog of Author Tim Ferriss
MongoDB | Blog
MongoDB | Blog
云风的 BLOG
云风的 BLOG
A
About on SuperTechFans
罗磊的独立博客
大猫的无限游戏
大猫的无限游戏
IT之家
IT之家

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
I built a 1:1 solar system with Fable 5 in the days befor...
Eric He · 2026-06-16 · via DEV Community

I built a 1:1 solar system with Fable 5 in the days before it was pulled offline

On June 9, Anthropic released Claude Fable 5 — the most capable model they'd ever made publicly available. I used it to build the core of a 1:1, real-time solar system that runs in a browser.

Three days later, on June 12–13, the US government ordered the model pulled offline worldwide over export controls. The tool that built this is gone.

I finished the rest with older Claude models, added full mobile support, and shipped v2.0.0. You can open it on your phone right now:

👉 https://sw.icodestar.net — pinch to fly from Earth orbit down to standing on the Moon.

The source and the entire (short) git history are public: https://github.com/hyqzz/Solar-Wanderer


What it actually is

It's not a skybox with planet textures. It's the solar system at true 1:1 kilometer scale, with every planet placed where NASA JPL ephemerides say it is right now:

  • Real positions. Planets are within 0.074° of JPL Horizons; the Moon within ~0.12°; 21 moons fitted from Horizons state vectors stay within 0.22° ten days out. Change the system clock and everything moves like the real thing — you can run time forward a decade at 10 years/second and watch the moons wind around their planets.
  • Seamless, no loading screens. Orbit → atmosphere → surface → walking → underwater is one continuous transition. Pinch down toward Earth and you fall through the atmosphere and land on the ocean; the water has buoyancy and you can dive.
  • The whole thing. 8 planets, the Moon, 21 fitted moons, the asteroid belt, the Kuiper belt, 28 real trans-Neptunian objects, 4 comets with anti-solar tails, and 21 real stars placed at their true 3D distances (they show parallax as you move). Out to 100,000 AU — the Oort Cloud, about 1.6 light-years from the Sun in every direction.
  • Tiny. ~200 kB gzipped, no backend, no account, MIT licensed. It's a static site.

Here's the part that still feels slightly unreal to me: you can pinch from Earth orbit straight down until you're standing on the ocean, watching the sun glint off the water — and it's the same continuous zoom the whole way.


The three hard problems

If you've tried to render space at real scale, you already know the enemy: floating-point precision. Three problems had to be solved together.

1. You cannot fit the solar system into float32

A GPU works in 32-bit floats. The distance from the Sun to the Oort Cloud is ~1.5 × 10¹³ km. Float32 has ~7 significant digits. If you store positions in absolute coordinates, by the time you're standing on a planet the per-vertex quantization is kilometers — the ground visibly shudders and shears.

The fix is a floating origin. All astronomy is computed in Float64 (heliocentric ecliptic J2000, in km), but before anything reaches the GPU, the camera's own position is subtracted out, so the scene is always centered on the viewer and the numbers handed to WebGL are small. The world quietly re-bases as you fly. Terrain noise is evaluated in tile-local coordinates for the same reason — sample noise at planet-radius magnitudes and fp32 banding wrecks it.

2. One depth buffer from 0.5 m to 10¹³ km

A normal perspective depth buffer can't span "half a meter in front of my feet" and "Neptune's orbit" at once — you get catastrophic z-fighting. The answer is a logarithmic depth buffer, which redistributes precision so near and absurdly-far geometry coexist. camera.far is set to 1 × 10¹⁵ km — about 106 light-years — so the 21 real background stars sit at honest distances and shift with parallax.

3. Making the landing seamless

The thing people feel but can't name is that there's no "enter planet" button. Orbiting, descending, atmospheric entry, touchdown, walking, and diving are not separate scenes — they're one camera following a single continuous dist-to-surface curve. On desktop the scroll wheel drives that curve; on mobile, the pinch gesture is mapped onto the exact same curve, so landing with a finger feels identical to landing with a wheel. The atmosphere is ray-marched (Rayleigh + Mie, Nishita-style) per planet, so the sky reddens correctly as you sink into it, and gas giants let you fly down into the cloud layers.


Try it

Pinch out until Earth is a pale blue dot. Then keep going.

Built with Three.js + WebGL2. MIT licensed. No account, no backend, no tracking.