What's New on DubVoice.ai: 3 New Voice Providers, Meta AI Video, YouTube Videos & UGC

A lot has shipped recently. Here is the short version, grouped by what it changes for you.

Text-to-speech: three new providers, 17,800+ voices

The voice library went from 10,500 to 17,800+ voices across six providers. The three new arrivals:

Vbee — 1,649 voices, 52 languages

Vietnamese-heavy (1,212 of the voices) but far broader than that: Arabic, Hindi, Japanese, French, Indonesian, Filipino and more. If you publish for South-East Asian audiences, this is the single biggest addition.

Fish Audio — 275 community voices

A multilingual community library spanning 15 languages, with 96 English voices plus Chinese, Spanish, Portuguese, Arabic and Turkish. Strong on character and personality voices rather than corporate neutrality.

Kokoro — open source, 70% off

Hexgrad's open-source Kokoro-82M, self-hosted. 1 character = 0.3 credits — the cheapest option on the platform after Edge TTS, with no rate limit and no text-length cap. 54 voices across 9 languages, and it supports weighted voice mixing (af_bella+af_sky(2)) which nothing else here does.

All three sit alongside ElevenLabs (15,190 voices), Minimax (701) and Edge TTS (90% off). Every one of them is reachable from the same dashboard and the same [public API](/dashboard/api-docs) — Vbee and Fish Audio synthesise through the standard TTS endpoint using a prefixed voice id.

The voice library itself was rebuilt too: proper cards with avatars, full metadata tags, per-provider filters (language, gender, age, region, category) and working previews for every provider.

Video: Meta AI at 2,000 credits

Video went from 3 models to 6. Two are new:

• Meta AI — 2,000 credits per clip. By a wide margin the cheapest way to generate video here. Supports square 1:1 alongside portrait and landscape, and does image-to-video with a start frame plus an optional end frame, interpolating between the two.
• Omni Flash — from 4,688 credits. The fastest model (30-90s), priced by duration, and it accepts up to 7 reference images — more visual context than anything else on the platform.

Prices moved on the existing family too: Veo 3.1 Fast halved to 7,500 credits. It is now the best value in the Veo range and the one to reach for by default — and the Veo family remains the only one that generates native audio.

Images: Nano Banana 2 Lite and Meta AI

The image picker is now 6 models. New:

• Nano Banana 2 Lite — 500 credits. The fastest and cheapest image model, for iterating on a prompt before committing.
• Meta AI — 1,500 credits. Uses named reference components rather than a flat list: character, scene and style images that bind to specific roles in the composition.

Nano Banana Pro remains the quality pick with 4K upscale, and Nano Banana 2.1 sits in the middle at 1,000.

YouTube Videos

The YouTube Videos tool takes a topic and produces a finished narrated video — script, voiceover and visuals — without hopping between tools. Useful for faceless channels and for turning an existing article into a video without rewriting it yourself.

UGC Studio — now live

UGC Studio has shipped: user-generated-content style clips, the format that carries most short-form ad creative right now. Pick one of 20 AI creators (or upload your own), write the script, attach a product photo, and the script is split into clips, filmed and stitched into one video — with lip-sync generated alongside the footage rather than dubbed on afterwards.

The same screen also does brand images: upload 1-5 reference photos of your product and get back a batch of on-brand shots — clean packshots, review-style photos, brand scenes or lifestyle frames — at 1,000 credits each. Full write-up: [UGC Studio guide](/blog/ugc-studio-ai-video-ads-brand-images).

What this changes if you were already here

A release note is only useful if it tells you what to do differently, so here is the short version for existing users.

If you write for South-East Asian audiences, the voice library just stopped being the bottleneck. Vietnamese in particular went from a handful of options to over a thousand, and the surrounding languages — Indonesian, Filipino, Thai — went from thin to usable.

If you render at volume, look at Kokoro before your next batch. At 0.3 credits per character it is the cheapest tier on the platform after Edge TTS, with no rate limit and no length cap. For draft passes and internal content the quality difference does not justify paying ten times more.

If you were avoiding video on cost, the floor moved. Meta AI at 2,000 credits a clip is cheap enough to use for thinking rather than only for finals, and Veo 3.1 Fast halving to 7,500 makes the tier with native audio the default rather than the splurge.

The thread running through all of it

Three of these changes point the same way: the cheap tier of each category got much cheaper, while the quality tier stayed where it was.

That widens the gap between drafting and finishing, and the way to use it is to stop doing both on the same model. Iterate on Nano Banana 2 Lite, Kokoro and Meta AI until the idea is right; finish on Nano Banana Pro, ElevenLabs and Veo 3.1. Teams that adopt that split usually find their spend goes down and their output goes up, because the expensive render only happens once.

What is not here

Two things worth saying plainly, because release notes tend to omit them.

Lip sync on generated video is still not solved — the Veo family generates speech, but a dubbed clip does not re-time syllables to a mouth. And the voice count is a library figure, not a quality guarantee: a 17,800-voice catalogue contains a lot of voices you would never use, and the filters matter more than the total.

For the full picture of what the platform covers rather than what changed, see [everything you can do with DubVoice.ai](/blog/everything-you-can-do-with-dubvoice-ai).

Also worth knowing

• Voice cloning now handles the full 10MB / 5-minute sample the interface always advertised. Larger uploads used to fail; they go straight to storage now instead of through the API.
• Kokoro discount deepened to 70% (was 50%).
• The public API covers Vbee and Fish Audio, with the full model line-up documented at [/dashboard/api-docs](/dashboard/api-docs) and in machine-readable form at [/llms.txt](/llms.txt).

Everything above runs off the same credit balance, starting at $4.99. No subscription, and credits never expire.