The method, with nothing hidden.

Step 1: Generate every legal splice

From two names, the engine builds candidates six ways:

• Prefix + suffix, both orders — the classic cut: every starting chunk of one name joined to every ending chunk of the other (the Brangelina family).
• Prefix + prefix — Tom + Kat. Short-name pairs live here.
• Suffix + suffix — rare, but occasionally the back halves are the famous parts.
• Overlap merge — where the seam letter matches, it collapses instead of doubling.
• First-syllable mash — the opening syllable of each name, in both orders.
• Initial fusion & full joins — one initial plus the whole other name, and the two complete names side by side.

Accents are folded to their base letters (José → jose) because that's how the blend would be typed as a handle. Anything without a vowel, shorter than 3 or longer than 15 letters is discarded.

Step 2: Score for readability

Each surviving candidate is scored on signals you can verify by eye:

• Vowel balance — scores peak when roughly 35–50% of letters are vowels. All-consonant results are unsayable; all-vowel results are mush.
• Length — 5–9 letters scores best. That's the hashtag-and-handle sweet spot, not an aesthetic opinion.
• Consonant clusters — a run of three consonants costs points, four costs more. This is where most splices die.
• Recognizability — a surviving chunk of each name (ideally 3–4 letters) earns points; a blend dominated by one name is penalized as a near-typo of it.
• Seam quality — cuts next to vowels earn a bonus; doubled letters are allowed once (Bennifer) and penalized after that; awkward doubled vowels (aa, ii, uu) cost points.

The three style buttons re-weight the same list: Couple takes the raw score, Ship boosts short and punchy, Playful boosts longer results, and the baby combiner adds a bonus for soft, vowel-ending, name-shaped results.

What the score cannot judge

• Meaning. A top-scoring blend might be an existing word — sweet, rude or taken. Search your favorite before printing it.
• Availability. No username, hashtag or trademark checks happen here; the generator has no network calls to check with.
• Non-Latin scripts. Blending is letter surgery, so names written in Chinese, Arabic, Cyrillic and other scripts can't be cut. The tool says so and suggests the romanized spelling instead of failing silently.
• Taste. The score ranks pronounceability, not romance. Couples regularly favorite the #7 result because it made them laugh — that's the correct outcome, and it's why favorites exist.

Short names deserve a footnote: two-letter names permit almost no splices, so when the honest candidate list runs dry the generator adds declared fallback forms (initial + name, both names together, paired diminutives) rather than padding the list with garbage. You can see exactly this behavior with Al + Jo in the generator.