AI Film Index

Casting the WPR/AIFI voice

For Stefan, Daphne and YZ: fifteen narrators reading the same story from our TikTok format. Sound on, compare, then reply to Dennis with a number.

⚙ Energy is a dial, not a choice — same voice (D-A1), three settings

You liked D-A3's enthusiasm but D-A1's cleanliness. Good news: enthusiasm is a render-time setting. Here's D-A1 at three energy levels — pick the timbre from the shortlist above, and the energy is tuned per video.

♪

D-A1 · energetic
low stability — A3's spark on A1's clean voice
♪

D-A1 · balanced
mid stability — the likely house default
♪

D-A1 · measured
high stability — cleanest, most formal

Round 3 — Designed voices · nobody else has these · 2026-07-12

Nine one-of-a-kind voices generated from three briefs. Unlike E1–E6 these are proprietary: pick one and it becomes the Index voice, unavailable to anyone else. Reply with e.g. D-A2.

D-A1

Designed D-A1ElevenLabs · one of a kind

Brief A — the curator: male 40-50, low warm, wry, close-mic, measured.
D-A2

Designed D-A2ElevenLabs · one of a kind

Brief A — the curator: male 40-50, low warm, wry, close-mic, measured.
D-A3

Designed D-A3ElevenLabs · one of a kind

Brief A — the curator: male 40-50, low warm, wry, close-mic, measured.
D-B1

Designed D-B1ElevenLabs · one of a kind

Brief B — the veteran: male 55-65, slightly gravelly, slow gravitas, dry humor.
D-B2

Designed D-B2ElevenLabs · one of a kind

Brief B — the veteran: male 55-65, slightly gravelly, slow gravitas, dry humor.
D-B3

Designed D-B3ElevenLabs · one of a kind

Brief B — the veteran: male 55-65, slightly gravelly, slow gravitas, dry humor.
D-C1

Designed D-C1ElevenLabs · one of a kind

Brief C — the young critic: male 28-35, soft-spoken podcast intimacy, cool precision.
D-C2

Designed D-C2ElevenLabs · one of a kind

Brief C — the young critic: male 28-35, soft-spoken podcast intimacy, cool precision.
D-C3

Designed D-C3ElevenLabs · one of a kind

Brief C — the young critic: male 28-35, soft-spoken podcast intimacy, cool precision.

Round 2 — ElevenLabs test · 2026-07-12

Is the paid voice clearly better than Andrew? Same line as Round 1. Reply to Dennis with E1–E6 or "keep Andrew". One ElevenLabs month covers the whole quarter.

A

Andrewcurrent pick · freecurrent voice

The edge-tts baseline the 45 videos use today. Zero cost.
E1

GeorgeElevenLabs · British

Warm, captivating storyteller. The classic documentary register.
E2

BrianElevenLabs · American

Deep, resonant, comforting. Closest to a trailer-narrator Andrew.
E3

DanielElevenLabs · British

Steady broadcaster. BBC-bulletin authority.
E4

BillElevenLabs · American

Wise, mature, balanced. Older-narrator gravitas.
E5

EricElevenLabs · American

Smooth and trustworthy. The most neutral of the six.
E6

LilyElevenLabs · British

Velvety actress. The female counterpoint.

★ The shortlist — 4 finalists · pick one

These four survived acoustic analysis + a human-naturalness model + a tuning loop — all top-tier, near-identical quality (MOS 3.58–3.66). The choice now is pure taste. Everything below is the fuller field for reference. Reply to Dennis with 1, 2, 3 or 4.

1

D-A1 · the curator

warm, expressive, brand-fit — Claude's pick
108 Hz · 135 wpm · MOS 3.58
2

D-C1 · the critic

cleanest & most natural, flatter/neutral read
106 Hz · 143 wpm · MOS 3.66
3

D-B1 · the veteran

more age and gravitas, slowest delivery
114 Hz · 131 wpm · MOS 3.59
4

V5 · tuned storyteller

auto-generated challenger, brand-pocket + clean
104 Hz · 136 wpm · MOS 3.64

Round 1 (settled: Andrew) — every voice reads this line

“Every February, the Berlin Film Festival awards its jury prizes to some of the best films in the world. This year, one went to a film with no camera, no set, and no cast. The Thread was made entirely with AI. And the jury knew exactly what it was.”

Ordered by Claude’s ranking for this job (natural delivery first, authority second, gimmick last) — gold numbers are its top three. It’s a starting point; trust your ears.

1

AndrewAmerican male

The most natural prosody of the fifteen. Warm, close-mic, modern. Disappears into the story.
2

BrianAmerican male

Relaxed, conversational. The closest to how good TikTok video-essays actually sound.
3

ChristopherAmerican male

Deep, authoritative, trailer weight. Strongest 'the Index has spoken' energy; slightly announcer-ish.
4

EmmaAmerican female

The most natural female option. Warm and effortless, reads like a person, not a product.
5

GuyAmerican male

Classic documentary narrator. Steady and clean, though widely used elsewhere.
6

RyanBritish malecurrent voice

Measured, slightly formal. What the current videos use.
7

AvaAmerican female

Bright, lifelike, a touch more energetic than Emma.
8

ThomasBritish male

Refined BBC-arts register. More precise than Ryan, less warm.
9

SteffanAmerican male

Even, measured narrator. Safe and clear, less character.
10

SoniaBritish female

Elegant and calm, gallery-voiceover feel.
11

AriaAmerican female

Polished broadcast tone. Crisp, but the most 'assistant-like' of the female voices.
12

ConnorIrish male

Distinctive lilt, the most character in the lineup. Memorable but the accent leads.
13

NatashaAustralian female

Natural and friendly. The accent reads casual rather than authoritative.
14

RogerAmerican male

Older, gravelly texture. Distinct, but rougher through phone speakers.
15

WilliamAustralian male

Laid-back Aussie. Likeable, least 'index of record' of the fifteen.

All fifteen are production-ready neural voices at zero cost — the pick swaps into the pipeline with one command and an overnight re-render of the video bank. If none of them lands, the upgrade path is a custom ElevenLabs voice (~$5–22/mo), same swap.