Labs · Free tools
Browser tools for readers
and card players
25 free tools. Most run locally in your browser: no text, no card file ever leaves your machine. A few static-data pages (the price table, the weekly card digest) are refreshed by an automated pipeline and labeled as such. The rules and parsers are the same ones our app pipeline actually uses — not demos.
Checkups
AI Flavor Checkup
Paste a Chinese passage; a phrase blacklist plus sentence-pattern quotas return a 0-5 score with every hit highlighted. Browser port of the v10.7 ruleset behind our card copy review.
Open tool →Card Checkup
Drop a character card (PNG / JSON / .charx): format version, field contents, token distribution, embedded lorebook, and the usual import-failure causes — in one pass.
Open tool →Book File Checkup
Drop a novel txt: chapter detection, word count, encoding, and audiobook length at four speeds — a 30MB tome parses locally with a real progress bar.
Open tool →Fiction Prose Stats
Sentence rhythm, dialogue ratio and verbatim repeats at a glance — dialogue bands referenced against 378 real chapter samples.
Open tool →Context Window Lens
See exactly why your AI chat forgets: drag the window slider and watch turns slide out — or drop your own tavern .jsonl and run it on a real chat.
Open tool →Tavern Preset Viewer
Turn a preset JSON into a readable config sheet: every prompt block in injection order, sampling parameters checked against common defaults.
Open tool →Tavern debugging
Lorebook Trigger Simulator
Paste a chat and see which world info entries fire and why: keyword hits, constants and recursion chains all drawn out.
Open tool →ST Regex Tester
Debug SillyTavern regex scripts online with a live before/after; catastrophic-backtracking patterns get caught by the same guard the app ships.
Open tool →ST Macro Playground
See exactly what {{char}} and {{random}} expand to — with a macro support table checked line-by-line against the actual engine.
Open tool →Greeting Preview
See your first_mes as a real chat bubble before you publish — and find out whether your 800-word opener arrives as 7 rapid-fire texts.
Open tool →Cards & file tools
Card Studio
Create a card from scratch or drop a PNG / JSON / .charx: per-field editing with live token counts, dead lorebook entries flagged as you type, macro preview, v2/v3 JSON or PNG export — edits auto-save locally.
Open tool →PNG Card Embed & Extract
Convert between JSON and PNG character cards: drop a PNG to pull the raw JSON, or a JSON plus a base image to build a PNG card byte-by-byte.
Open tool →Card Format Converter
Convert v1 / v2 / v3 cards — downgrades list every dropped field and ask before download, with a three-generation comparison table.
Open tool →TXT to EPUB Converter
Turn a txt into a real ebook: production-grade chapter detection, GBK rescue, a canvas-typeset cover, packaged as EPUB 3 — passes EPUBCheck with zero errors.
Open tool →Image Vectorizer
Convert PNG / JPG / WebP into clean SVG vectors: the open-source VTracer engine as WebAssembly, four presets, slide-to-compare — no watermark, unlimited.
Open tool →Cost & pricing
AI Reading Cost Calculator
What a 2M-word audiobook actually costs, in real prices — 19 mainstream models at list prices, with a “what 5,000 sign-up credits buy” comparison.
Open tool →LLM API Price Table
1,300+ models from 30 providers in one sortable table: input, output, cache-read price and context window.
Open tool →Prompt Cache Savings Calculator
How much caching saves on long chats, at real price spreads — hit-rate anchors from our instrumented four-way benchmark.
Open tool →Play & share
Turing Arena — Human or AI?
One blind question a day: two fiction passages side by side — spot the machine's output. In our double-blind study LLM judges scored 12%; see if you can beat them. Wordle-style scorecard, streaks stored locally.
Open tool →Fan-Card Maker
Frame a favorite character as a collectible card or bookmark, drawn on canvas and downloaded as PNG; rarity is the card's real community standing.
Open tool →Try Your Book
Drop a txt into a web phone simulator and read it with real pagination and page turns — first three chapters sample locally.
Open tool →Chat Log to Image
Drop a tavern .jsonl or paste "Name: text" lines, pick the range worth showing, and get one typeset long image — four themes, auto-paged into a zip for long chats.
Open tool →Content library
New Cards Weekly
This week's community character cards on one page: cover, tags, lorebook size and a one-line hook, each linking to a detail page.
Open tool →TTS Voice Gallery
96 voices reading the same sentence: Qwen-TTS (49, incl. dialects), Grok Voice (26), OpenAI (13), CosyVoice2 (8) — loudness-normalized, loaded only when you press play.
Open tool →Experimental tools
Tech previews: working end to end, with honest limits documented on each page. Results vary by input — read the on-page “what it can't do” section first.
Resource pages from the same toolbox
Not interactive tools, but free and built from the same production assets.
Official prompt template library
20+ official prompt packs on one page: continuation styles, story-option directors, video and image wrappers — full text, copy-ready, with a detail page per template.
Foreverse Research
Benchmarks you can recompute: the nine-model continuation leaderboard, the four-way tavern comparison, and the citable research archive — raw chains and JSON snapshots included.
Fiction Bench — continuation leaderboard
Nine models, two genres, 20 consecutive rounds each, double-blind judging: which LLM actually writes fiction like your book, with failure taxonomy and cost per 10k characters.
Foreverse vs SillyTavern benchmark
Same model, same key, byte-identical scripts, one metering gateway: measured cache hit rates, cost, deep-lore recall, and long-chat memory across four setups.
What these tools don't do
They don't store your input, so there's no “smarter in the cloud” tier; and they aren't truth oracles — the flavor checkup flags known patterns, not authorship, and the card checkup reports format facts, not quality scores. Each tool page has a “what it can't catch” section; read it before trusting a result.