InVideo AI Voice Cloning: AI Twins, Agent Voices and TTS Credits

InVideo AI can clone your voice in two places, and neither one is the screen you land on. The classic route sits in AI twins, behind the Older versions button: you upload a short, clean recording, read a permission statement and finalize the clone, which invideo says takes a few minutes ^[1]. The newer route runs through Agent Two, where the Seed Audio 1.0 model clones a voice from one to three short reference clips. Both routes spend credits. Use the classic flow for a quick narrated stock video, and the agent when the voice has to live on an editable timeline.

Two routes to a cloned voice, and where each one sits

The prompt page now opens Agent Two by default. The original InVideo AI flow, where you paste a script or prompt, get a stock-media video with AI voiceover and subtitles and then edit it with Magic Box commands, is still reachable through the Older versions button on that page; our text to video guide walks through it. The help center places both the avatar library and AI twins there, so anyone searching the default agent screen for a voice-clone menu will not find it. We found no date on how long the older versions stay available, so treat the route as current, not permanent.

An AI twin is invideo's term for a realistic digital version of you that speaks, gestures and presents from a typed prompt, and the voice clone is its audio half. Agent Two splits the same job across third-party models instead. A voice model clones or speaks the script, and a talking-avatar model such as HeyGen Photo Avatar, Kling Avatar or OmniHuman 2.0 turns a presenter image and that audio into one clip on the timeline, priced per second. Changing what that avatar says means generating new footage, so settle the script before you render.

Cloning your voice in AI twins, step by step

The classic flow is short, and the help center documents every step of it:

1. On the prompt page, open Older versions, then the AI twins section.
2. Upload a short, clean recording of your voice as an MP3, WAV or M4A file.
3. Read the permission statement the flow asks for, which records the speaker's consent.
4. Finalize the clone and wait a few minutes while invideo builds it.
5. Select, rename, delete or set the clone as your default from the AI twins menu.
6. To put the clone on a video you already generated, tell Magic Box to change the narrator voice.

The classic editor has a few voice limits worth knowing before a long project. You can change voiceover speed in the Script panel or with a Magic Box pace command, but on an avatar the tone cannot change, and a speed change alters the video's length and therefore its credits. Magic Box cannot mute the voiceover in only one part of a video. Pauses go in the Script panel, where Add gap here inserts a break and stretches the media. Stock voices named in the prompt cannot be previewed before generation.

Cloning in Agent Two with Seed Audio and ElevenLabs voices

In Agent Two you hand the agent one to three short, clean clips of the speaker and tag your lines in the prompt, and it clones the voice with the Seed Audio 1.0 model. For ElevenLabs cloning, invideo's model page says the agent first checks that the voice is yours, then plays the clone back for approval. If your workspace already holds ElevenLabs cloned voices, they appear in the ElevenLabs voice library inside invideo. One rule trips people up. A voice id works only with its own model, so a Seed voice cannot be called from ElevenLabs, and the docs advise picking the voice once per script.

Agent Two's default setting is to allow generation without asking, and you can switch it to ask before every generation, which suits voice work where retakes add up. Outside agents connected through the invideo MCP server get a dry-run price first and wait for your yes on anything costing one credit or more. Some voice models also decline to copy a voice without consent, so a refused take may be the model's own rule rather than an error.

Language is where a clone pays off. invideo says a cloned ElevenLabs voice is not tied to the language it was recorded in and can speak the 29 languages of Eleven Multilingual v2, and that the agent can translate your script for the clone to read. Dubbing still runs step by step, because no single model dubs a whole video: the agent transcribes, translates, voices the new script and re-syncs lips with a lip-sync model where a mouth shows. invideo also says cloning your own voice sits in YouTube's safe zone, while near-identical text to speech over stock footage is risky; our faceless YouTube guide covers that workflow.

Text to speech when you do not need your own voice

Most narration does not need a clone. The model reference lists ElevenLabs TTS v4 and Multilingual v2, ElevenLabs Text to Dialogue, Seed TTS 2.0, MiniMax Speech 2.8, Cartesia Sonic 3.6, Gemini 3.1 Flash TTS, Fish Audio S2.1 Pro and Sarvam Bulbul v3 ^[2]. They take different controls. ElevenLabs v4 reads emotion tags in the text, Seed TTS 2.0 offers speech rate and silence duration, and Gemini 3.1 Flash TTS takes a second speaker for dialogue. invideo gives 50+ languages as its own figure for AI voices, subtitles and dubbing combined.

These rates come from invideo's audio pricing page ^[3], which advises cheaper models for temp tracks and premium ones for the final cut. The exact cost appears in the app before you confirm. invideo can change credit costs at any time without notice, and when a price is an estimate the final charge can end higher. Seed Audio 1.0 costs more per character than any stock voice in this table. That gap repeats on every retake.

Consent rules that apply before the first take

InVideo AI requires explicit consent from the person in the recording, captured in a separate in-product step that can be withdrawn; celebrities and politicians cannot be cloned, and an avatar of anyone under 18 is forbidden even with parental consent ^[4]. When cloning is on, invideo may run automated liveness or consent checks, and its celebrity detection is automated too, with human review on request. The dubbing guides go further. They ask for the speaker's consent in writing before you clone an actor or any other speaker.

The broader rules follow from that. The acceptable use policy bans misleading deepfakes of real people, you must hold consents from anyone identifiable in your outputs, and the help center says to clone only a voice you have the right to use. Human avatar actors cannot be used for alcohol, tobacco, cannabis, crypto, gambling, MLM, adult services, political commentary, religion or taking sides in a war. If you are weighing HeyGen for the same job, its Digital Twins start from about 15 seconds of footage plus an on-camera consent clip, and our HeyGen voice cloning guide covers that side.

What a cloned voice costs, and what survives a cancellation

Credits come with the plan, per seat. Starter costs $20 a seat a month billed monthly with 400 credits a seat a month. Plus costs $60 a seat a month billed monthly with 2,000 credits. Max costs $150 a seat a month billed monthly with 5,000 credits. On the annual card, Starter showed $10, Plus $30 and Max $75 a seat a month billed annually and paid upfront for the year, as an offer shown in November 2026. Credits reset monthly on annual billing too, never roll over, and Agent Two's own work spends them alongside the voice.

The Free plan lets you try voices, but voice generation uses credits against its weekly limit. The current help center does not say which plans include voice cloning, so check your model list before paying. Starter is the plan to watch: it gets Agent Two Lite only and selected models, and a model missing from your list means your plan does not include it. Voice is also personal data, so the training rule matters. On Free, Starter, Plus and Max the Terms let invideo train on your data; the Team and Scale cards promise no training, so get that in writing. Third-party models sit outside that promise: Standard Tier providers are barred from training on your content, Experimental Tier providers have made no such commitment.

Canceling moves the account to Free at the end of the paid period, and unused credits, including purchased extra credits, expire. A cloned avatar stays available afterward, generated media stays downloadable, and invideo keeps voice profiles until you ask for deletion or close the account. You can withdraw consent in account settings. Fees are non-refundable; invideo's Terms allow a refund request within 24 hours of a first purchase or renewal only if no credits or premium features were used, and credits, including extra credits, are never refunded. The pricing guide has the full terms.

The bottom line

InVideo AI handles voice cloning well enough for narration, but you have to know where to look. The AI twins flow behind Older versions is the quick path for a classic stock video, while Seed Audio in Agent Two suits timeline work, translated scripts and dubbing. Budget for retakes. A Seed Audio clone costs more per character than the stock voices, credits never roll over, and the consent step is mandatory. Record clean audio, get written consent from anyone who is not you, and confirm your plan lists the voice model before you commit.