luminarforge.
Pricing Compare Trust Log in Start free
The voice foundry — Voice DNA

Your client’s voice, cut like a typeface.

Three real writing samples in. A measurable profile out — ten style metrics with tolerances, plus extracted traits — and every draft is checked against it like a proof against its specimen. Not a vibe. A measurement.

Start free — 2,000 credits See the proof check ↓

Voice DNA on every plan · the specimen below is our own — measured, not mocked

Fig. 1 — voice specimen № V-001 Real profile — extracted by the production engine

Luminar Works — House Voice

Cut 2026 · from published copy
DisplayLegible by construction.
LedeThe fix is sequence, not quantity.
BodyAdd a client, paste their URL, and the same design comes back in their identity — correctly. A plate resolves the brand at render time, from whichever client workspace opened it.
Sample lines are our published site copy, verbatim · profiled from 3 samples · 4,129 words · by the production extractor — the same code that profiles your clients

Characteristics — measured

Avg sentence length16.32 WORDS
Sentence variance12.74 σ
Vocabulary complexity8.86 GRADE
Questions / 100 words0.12
Exclamation rate0 %
Contraction usage1.48 %
Sentence-opener diversity0.46
Emoji frequency0.02 /100W
Paragraph length (avg)35.29 WORDS

Plus extracted traits — formality (1–10), directness, humour — from the AI trait pass. Every metric carries a tolerance for the proof check.

Measured — not guessed
The proof check — same announcement, set two ways

Generic doesn’t fail on taste. It fails on measurement.

Below: one announcement, written twice, both drafts run through the production voice-match scorer against the specimen above. The readouts are its actual output. The generic draft isn’t wrong because it’s ugly — it’s wrong because it’s measurably not this voice.

The monotone is the tell: every sentence the same length.

Draft 1 — no profileGeneric
We’re thrilled to announce the launch of our brand-new schedule desk! In a world where content teams are busier than ever, staying organized is absolutely essential for success. Our innovative new solution empowers agencies to streamline their entire planning workflow and elevate their content strategy. With powerful campaign phases, intelligent scheduling capabilities, and comprehensive milestone tracking, teams can now manage everything from a single, unified surface…
Sentence variance3.39 vs 12.74 — MISS
Vocabulary complexity13.07 vs 8.86 — MISS
Exclamation rate33.3% vs 0% — MISS
Sentence length15.83 vs 16.32 — MATCH
Voice match65
Set in the house voice
Draft 2 — profile appliedVoice-set
The schedule desk is live. The month is planned the way agencies actually plan it — a campaign with dated phases, drawn as bands across the calendar, with the client calls pinned between the posts they gate. The phases are ramp-up, live and follow-up, and the database refuses to save them out of order. It answers one question. It tells you what ships this week, for every client workspace you run, without anyone calling a status meeting to find out…
Sentence variance9.25 vs 12.74 — CLOSE
Vocabulary complexity6.64 vs 8.86 — CLOSE
Exclamation rate0% vs 0% — MATCH
Opener diversity0.44 vs 0.46 — MATCH
Voice match93

What surprised us, dogfooding this: our first “punchy” rewrite scored 66 — worse than the generic draft. The measured voice wanted longer, more varied sentences than we’d have guessed. That gap between what you’d guess and what’s measured is the whole point of the instrument.

Readouts: production scorer, actual output · tolerances per metric
The foundry — how a voice gets cut
Day one — Auto Tone DNANo questionnaires

A working voice before you ask for one

The moment a client’s knowledge base exists, a tone profile is generated silently from it — three descriptors, vocabulary, sentence style. No twenty-question onboarding. It’s the default the writing room uses until you cut the full specimen.

Generated from the brand brain · day-one default
The full cut — Voice DNA3 samples in

Three real samples, one measured voice

Paste three pieces of the client’s real writing — 2,000 words minimum, the same bar we held ourselves to above — and the extractor cuts the full specimen: ten quantitative metrics plus AI-extracted traits. The Writer drafts in it, the Humaniser polishes against it, and the proof check scores every draft.

How the writing room uses it →
Per client workspace · re-cut any time the voice evolves
Said plainly — how we keep this honest
On what “sounds like you” means here

It means named metrics with tolerances, measured deterministically — the numbers on this page are the instrument’s own. The trait layer (formality, directness, humour) is an AI reading on top, and we say so. No soul-absorption claims.

On this page’s specimen

The profile above is ours, for real — cut from our own published copy by the production extractor, before/after scored by the production matcher. When it surprised us, we printed the surprise.

The foundry is open

Cut your first client’s voice.

Start free with 2,000 credits — paste a client’s URL for a day-one tone profile, or bring three real samples and cut the full specimen. Roughly a week of real production, on us.

Start with 2,000 free credits See the rate card

No card · No trial countdown · No per-client fees

luminarforge.
LinkedIn-native today · Instagram & Facebook rolling out
Features
AI Pipeline Voice DNA Campaigns Templates AI Imagery Publishing SEO Hub AI Suggestions
This page
The proof check How it’s cut The rate card Compare
Catalogue
The print room Pricing Trust & security Start free Log in
Company & legal
Contact Terms Privacy Cookies

The specimen, characteristics table and voice-match readouts on this page are the production engine’s real output on Luminar Works’ own published writing (3 samples, 4,129 words). The two exhibit drafts were set for this page and measured by the production scorer. Trait extraction (formality, directness, humour) is an AI pass and is labelled as such. © 2026 Luminar Works.