The voice route

Every final line comes from ElevenLabs Eleven v4 on your own account. A native voice artist is the fallback when the AI voice can't reach the bar, never a route chosen ahead of it.

What the evidence says. English on ElevenLabs is [proven] on earlier versions and [official guidance] on v4, because no creator has released v4 work yet. Arabic on v4 is [best available, not yet validated]: no released Arabic production credits ElevenLabs, and no independent v4 dialect comparison exists. So each role is cast by ear, dialect by dialect, and a native artist stands behind it.

The route for every final take is your own ElevenLabs account, on a paid plan. Voice runs there only (decided).

• Use it through the full API (the interface ElevenLabs offers to programs), or through the web app with the same settings set by hand (steps in Voice and Arabic dialects → ElevenLabs settings, tags and pace → Do it yourself).

• It alone gives you the Voice Library's dialect voices, the language code, pronunciation dictionaries, Stability and lossless output. A free key refuses library voices.

• Not a master: any route that hides Stability or the output format. Higgsfield's built-in v4 is not used for voice: it offers preset or element voices only, with no Voice Library, no dictionaries and no language code.

• Model: Eleven v4 for every final line. v4 Turbo is made for real-time agents and is a different voice: never mix the two. Note the exact model ID on your first call, and keep it in your take log.

• Seedance never voices Arabic, crowd murmur (walla) included (decided). Every Arabic voice is recorded here. Arabic crowd murmur comes from real CC0 crowd recordings (see Sound design and mix → Real recordings from Openverse (CC0)) or from recorded voices; Arabic murmur in a take is muted and replaced (see Sound design and mix → Keeping or muting in-shot sound). No Arabic re-voiced read ships in any version.

• Seedance's own voice, outside Arabic only. Option: use it when a principal line is in English or another language Seedance speaks well, and you choose it for the project [proven]; the client approves the voice from a sample (see Speaking shots and lip sync → English lines and Seedance's own voice).

• Money: every batch of lines waits for an approved estimate (see Voice and Arabic dialects → Voice takes and the delivered file). Price voice in US dollars at the API rate. Where you work in the web app, it counts ElevenLabs credits from your plan: write those down as ElevenLabs credits. Never add either to Higgsfield credits.

• The plan itself (ElevenLabs pricing, read 19 Oct 2026): Starter at $6 a month, which carries the commercial licence; Creator at $22 a month. Check on first use which plan allows 48 kHz WAV (see Voice and Arabic dialects → Casting a voice).

The fallback: a native voice artist

A native voice artist [proven] is a fallback only (decided). Top productions where the acting carries the dialogue, Critterz, Primordial Soup's 1776 and László Gaál's work, recorded people. Bring a native artist in when:

• the dialect's best AI take fails the English-quality bar: the native listener rates it below 4 of 5 on authenticity, pronunciation or naturalness, or won't call it as good as your English ElevenLabs reads, or the back-check shows a wrong content word;

• an acting-critical line (shouting, crying, comic timing) still fails after the fix ladder (see Voice and Arabic dialects → Spelling for the engine);

• the role is a child.

Who, in this order:

1. the native reviewer, if they can voice it, after a paid sample read;

2. a remote native artist with a home studio from a voice marketplace, chosen on a paid sample read of the actual line;

3. a studio session in Cairo, Riyadh, Jeddah or Dubai, for live direction or a broadcast buyout.

What you prepare, and what you approve: first the brief, then the booking.

• Casting brief: dialect and sub-dialect, gender, age, register, character notes, the best AI draft as a timing guide.

• Script sheet: the approved line, vowel marks only where the reading is ambiguous, the English meaning, the emotion, the target length in seconds.

• Recording spec: dry WAV, 48 kHz, 24-bit, mono, one line per file; two or three readings at the timed length plus one freer read; 1 s of room tone.

• Session: directed live by you or the reviewer against the animatic (the storyboard frames held for their scenes' durations), or self-directed from notes with one pickup round.

Cost: about $25–80 (US dollars) per 30-second spot from marketplace artists, with a 24-hour turnaround; TV buyouts and studio sessions are quoted per project. Use ElevenLabs Voice Changer on the artist's read only if the cast timbre must differ, and judge the first converted line by ear before converting more. With no artist in time, the best AI take goes ahead with the native reviewer, marked draft until approved. With no reviewer either, Saudi, Gulf and every other non-Egyptian line stays a draft and never reaches the client (decided).

Sources

• ElevenLabs, What is Eleven v4? and the v4 prompting guide; Critterz, Primordial Soup's 1776 and László Gaál's films (human recorded voices; no links in the research).