10 seconds to clone a voice: ElevenLabs IVC vs cloning in Sume
ElevenLabs Instant Voice Clones start from 10 seconds of audio. Sume voice cloning is free of charge and sets a language per voice; here is how they differ.
ElevenLabs says Instant Voice Clones need as little as 10 seconds of audio, with Professional Voice Clones also supported on Eleven v4. In Sume, cloning a voice is not billed: the pricing policy lists voice clone among the things that stay free. The two differ less in price than in what the clone is bound to. A Sume voice carries a primary language that requests are checked against, and the sonic-preview model rejects pro voice clones.
In both cases, clone only a voice you have the speaker's clear permission to use. A short sample makes a convincing copy quickly, which is the reason consent has to come first.
What the pages state
| Item | ElevenLabs | Sume |
|---|---|---|
| Quick clone | Instant Voice Clone from 10 seconds of audio (Eleven v4 announcement) | Voice clone through the voices_create tool; minimum sample length not stated in the Sume docs |
| Higher-fidelity clone | Professional Voice Clone supported | Pro voice clones exist; sonic-preview rejects them with voice_model_mismatch |
| Latency note | Sept 28 changelog: a PVC can use its IVC version for lower latency | Not applicable: TTS is an async job |
| Price of cloning | Plan and credit based; not itemised on the pages read | Free (operations pricing policy, live since 2026-09-22) |
| Language | 90+ languages on v4 | Primary language stored on the voice; mismatch returns 409 tts_voice_language_mismatch |
Free to clone is not free to speak
A Sume clone costs nothing to create, but every TTS job that uses it is metered per character. On the Sonic engine that is Cartesia list x1.25 ($0.0475 per 1K characters from the $38 per 1M credits list), rounded up to the cent per job, plus the 5.5% platform fee. Cartesia's own pricing page gates instant cloning at its Pro plan and professional cloning at Startup, which is the vendor-side arrangement behind that engine.
So budget the speaking, not the cloning: a 1,000-character read is $0.0475 before fee, billed as $0.05 after rounding.
Picking between them
- Need a clone on an Eleven voice with v4 audio tags: use ElevenLabs. Sume does not ship Eleven.
- Need the clone to live next to your avatars, scripts and billing, on Sonic: clone it in Sume.
- Set the voice language when you create it. A wrong value only shows up later as a 409 when a request disagrees with it.
- Test with the exact lines you will ship. A 10-second sample tells you the voice is close, not that it holds up across a long script.
Sources
Related posts
More in Use cases
- Email header image 600x200: render 1920x640 at exactly 3:1
A 600x200 email header is 3:1, the widest shape GPT Image 2.5 accepts. Render 1920x640, then export 1200x400 for retina and 600x200 for 1x with Pillow.
- Employee handbook page to a 20 s training clip: one policy, one scene
Turn one handbook page into a Wan 3.0 training clip on Sume: pick the policy, picture the behavior, type the rule as captions, and review it with HR before use.
- Employee onboarding videos: ten 30-second Wan 3.0 modules for $37.50
Ten 30 second onboarding clips on Sume's Wan 3.0 cost $37.50 at 720p, $18.75 at 480p or $75.00 at 1080p. Concurrency by plan and what to keep out of frames.
- English podcast clip to a Spanish short: dub and captions, $0.38
An English podcast clip turned into a Spanish short costs up to $0.38 on Sume: trim, detach, transcribe, speak, render and one caption job at $0.20.
Written by Sume