Kyn Smith 🔨 — Roadmap
Where the tool is, and where it's going. Ships in order; the changelog (📋 on the main page) records what actually landed. Also: this page · back to the smithy
✅ v1.7 — shipped 2026-10-06
- Five SK tag boxes — Physical, Identity, Personality, Clothing, Location — matching Soulkyn's real editor layout. The old six-box system (pose/expression/lighting/composition) is replaced with the categories SK actually uses
- Per-box "Copy tags" — each box has its own copy button that copies checked tags as comma-separated values, ready to paste into the matching SK box
- Per-box add-tag input — type custom tags and hit Enter; 18-tag cap, 35-char limit, duplicate blocking, live count (N/18) per box
- Expressions → Physical — smile, smirk, blush, serious, teary eyes, etc. fold into the Physical box (SK reads facial expressions as physical tags) — 1–4 chips max
- Lighting/composition → Location — golden hour, from above, portrait framing, candlelight, neon, etc. fold into Location alongside setting tags. Physical, Clothing, and Location all feed SK's image gen
- Personality box — auto-filled from the vision model's personality array, matched against a 20-trait taxonomy for canonical formatting
- Export options — Copy sheet / Export .txt / Export .md / Send to imagegen, all from a dedicated save card
✅ v1.6 — shipped 2026-10-05
- Gender toggle (♀ Female / ♂ Male / 🎲 Random) — forces the sheet's gender regardless of the vision model's guess
- Gender-aware race tags and cross-gender tag stripping
- Analysis prompt rewritten with explicit gender detection instructions
✅ v1.5 — shipped 2026-10-03
- Auto-tagging — the analysis reads feed an SK-format tag matcher (built on the Danbooru catalog): canonical tags per box, confidence-ranked, with a why trace per chip
- Rules engine: ≤18 chips per box, ≤35 chars per tag, no cross-box duplicates; identity box stays suggestion-only, never auto-checked
- SK Tags block in the sheet — chips editable by hand, edits re-write the sheet block live
- Full-name prompt: the analysis demands a full name (first + last), never a stock default
- Changelog modal now data-driven (changelog-data.js)
✅ v1.4 — shipped 2026-10-02
- Outfit & background/location read from the image — the analysis returns an outfit paragraph and a location line, shown in the sheet and used in the preview prompt
- Preview image now shows the character in her outfit and setting
- → Send to imagegen carries outfit + location with the character — they land there as custom presets, preselected
- This roadmap page
🗺 Later — honest maybes
- Round-trip: imagegen generations analyzed back through the smith (regenerate → refine loops)
- Sheet editing after analysis (fix the vision model's reads before copying)
- Batch mode: a folder of images, one sheet each
- Personality & greeting depth controls (short / vivid / long-form)
- Persona versioning — every card edit saved as a version with seed & params, full history
- AI breeding — two cards in, one child card out: weighted tag inheritance + one mutation, lineage on the card
- Card → prompt writer — one tap turns a finished sheet into a weighted, engine-ready prompt
- Gacha persona roll — "roll me a stranger", built on the outfit-gacha tables
- Card gallery / deck view — every card on one shelf, TCG-style
🚧 Not planned
- Accounts, servers, or key storage — the key stays in your browser, same as always
- Auto-send without the button: nothing leaves this page unless you press → Send to imagegen