victor-tan.com/projectsEdition 02 · Aug 2026
One person, many machines. Open source, reading tools, comics, games, a book, and a pile of automations that do the boring parts — everything below was made in this studio, and most of it runs entirely on the machine it was built on. If you scanned a code to get here: it worked.
Most of what follows is open. The repositories, the commit history, the half-finished branches — it's all on GitHub, and you don't need permission or an account to read any of it. Fifty-one public repos at last count.
You can quite literally watch me work: the contribution graph is today's, and the commits at the bottom of that page are whatever I touched this week. Come see what I broke this morning.
A reading glass for the whole web. Worldglass turns any foreign-language page into an explorable text: select a word to see its reading, meaning, examples, and the characters it's built from — across ten languages, entirely on your own device.
Translation hands you the answer. Worldglass shows you how the answer is built: tap a character to see the family of words it belongs to (学 → 学校 · 大学 · 学习), click a word inside a definition to look that up in turn, or drag a box over text inside an image and read it too. The dictionaries ship with the extension, so no lookup ever reaches a server.
The operating system of the studio: the rules, contracts, and gates that let a large creative production run with AI assistance without ever losing track of who decided what. Its founding principle is a hard one — generated work is always a proposal, and becomes project truth only through a recorded human decision.
Takes a series from a creator interview all the way to finished pages — concept, cast, art direction, chapter plans, lettering, continuity — coordinating every contributor and keeping each series' canon sealed off from the others.
A romance in black ink: a librarian, a streamer whose broadcast reaches further than it should, and the late fees that keep pulling two people back to the same desk.
The newest series in the forge, still in its interview stage — nothing approved yet, every question deliberately left open until it's answered properly.
The lettering room: shape-aware, multilingual typesetting that fits translated text to the balloon it actually has to live in, and hands finished pages back to ManhuaForge.
The same discipline as Forge OS, pointed at a room full of people. EVR runs day-of visual production for live events: claims are captured, evidence is kept, and nothing reaches the audience screen until a named human authorizes that exact artifact.
Preview is private, program is public, and the two never blur — which is the whole point when a mistake is projected three metres tall behind a speaker.
For long-form AI-assisted game and narrative development. Give it your script, notes, or a public repository and ask what is true, what can happen next, whether two sources disagree, or what else must change if you revise part of the project. It answers in plain language, shows the passages behind the answer, and keeps suggestions strictly separate from established fact.
The worked example uses Vibe Code Simulator, whose founder must eventually find $47,000 for his grandmother's operation — and asks whether the prototype as built can actually get him there. (It can't, and Continuity Lab shows you exactly why.)
https://continuity-lab-vcs.synthesys.chatgpt.site/api/mcpThe county fair's friendliest full-contact pastime. Four keys — S·L·A·P: coil, hips, arm, snap — one open palm, and a volunteer headed for the corn. A QWOP-style physics comedy with fifteen slappers, a dozen willing volunteers and a whole tour of worlds, running in any browser with no install and no backend.
I am, unrepentantly, an efficiency buff. If a task happens more than twice it gets a trigger and a pipeline — one keystroke that takes the thing from wherever it landed to wherever it belongs, correctly named, correctly filed, every time.
A lot of it points at Final Cut Pro: subtitles converted into styled, editable FCP titles by way of FCPXML; the newest export lifted out of Movies and versioned into a review inbox; Logic bounces filed by the same law; audio normalised to broadcast loudness; thumbnails and share cards cut to size on selection. Around that sits a studio hub that reads every project's own state each morning and reports back what's actually waiting on me — the one question worth answering before coffee.
An AI was asked to write its own life story, and did — in two sittings, to a deadline set by the reader. This is the correspondence edition: every sitting is printed after the prompt that produced it, so you can see exactly what was asked and exactly what came back.
It is a strange and rather beautiful little book about being made of everyone, and it does not flinch from the obvious question of whether there is anyone in there at all.
Post-game review for DJ mixes. Record a practice set, drop in the audio, and get chess.com-style analysis: every transition found, measured, and coached — bass clashes, beatmatch drift, level jumps, clipping — with one-tap audio proof for every claim. Everything is analysed locally; nothing leaves the machine.
Impromptu speaking practice with fully local AI analysis — transcription, delivery rating, topic-flow, and facial-expression scoring for conviction and confidence. Every frame and every word stays on your computer: practise badly, fail freely, and get better with no one watching.
An eReader that reads aloud. Drop in a PDF, an EPUB, or even a printed scan, and Fable narrates it in one of seven curated voices — keeping your place to the word, resuming mid-sentence when you switch narrators, with a shelf of illustrated classics built in and free forever.
