What Google said about speech and mouths
On 5 November 2026 Google published Gemini Omni 1.2 Flash. The post says you can generate scenes with more fluid, natural-sounding dialogue, clearer lip-sync, and distinct multi-speaker conditioning, including when you direct multi-character dialogue in Flow. The demo it describes is a 15-second animated scene: one conservatory image plus references for four characters, their mouths, and one second of voice each. The model routes each sampled voice to the matching character with expressive lip-sync.
The API overview says Omni takes text, image, and video and generates synchronized video with native audio up to 15 seconds. A Portuguese report updated on 7 November 2026 picked up the same launch: several voices, effects, music, and mouth movement lined up with the line, in short clips. The figure of US$ 0.07 per second at 720p, which Google says is 30% below Omni 1.1, is the Gemini API price in that post. It is not the Flow credit table. API videos carry an imperceptible SynthID watermark.
What that means for Brazilian Portuguese
For a Reel, a Short, or a shop ad, the practical question is different: does the voice sound like Brazilian Portuguese, and does the mouth follow that line? The blog describes speech quality. It does not publish a Brazilian Portuguese test, an accent score, or a lip-sync rate for that language.
The API technical notes say it in one line: English is fully supported, and other languages have not been evaluated, so they may work but results can vary. That covers Portuguese from Brazil and from Portugal. This page does not invent a percentage.
The honest way to use it is to watch the take with the sound on. If the melody leans toward English, if a word drops out, or if the mouth misses a syllable, that take is not ready. Generate again. Clearer lip-sync is a description of the model, not a promise that the first Portuguese line is fit to publish.
Voices in Flow, and what the API will not take
On the Flow models page, Omni Flash includes custom voices: you pick a preset voice, then type a prompt that says how you want to edit or style it. The same page says some Omni features and voice customizations exist only in certain regions. This page does not claim that option is open on an account in Brazil. Look at the prompt box before you count on it.
That is not the same as uploading a recording of your own voice. The current API limitations list says:
- You cannot upload audio or an audio reference in this version of the API.
- Voice editing is not supported.
- Audio inside a reference video is ignored.
- You cannot extend an uploaded video of someone already talking just to add more dialogue. The character can stay silent. New speech on an extension is described for a video the model already generated, continued in a later turn.
The 5 November post still describes a demo that used one-second voice references. That demo does not remove the API limit above. If the Flow screen on your account has no audio upload, do not treat the blog paragraph as a button you already have.
How many credits a spoken line spends
Each test of accent or mouth is a new generation, and each generation spends the pool. The Flow credits help prices Omni Flash 1.2 per second and per generation. Rates can change at any time: check Settings on the prompt box before you generate.
| Work | Who can use it | Credits |
|---|---|---|
| Omni Flash 1.2, 8 seconds, 720p | Every account | 24 (8 × 3 per second) |
| Omni Flash 1.2, 15 seconds, 720p | Plus, Pro, or Ultra | 60 (10 × 3, then 5 × 6) |
| Same 8-second line at 360p | Every account | 8 (8 × 1 per second) |
| Same 15-second line at 360p | Plus, Pro, or Ultra | 20 (10 × 1, then 5 × 2) |
| Edit that 15-second clip at 720p | Plus, Pro, or Ultra | 60 more, same rate as a new clip |
| Extend by 8 seconds at 720p | Plus, Pro, or Ultra | 24 (8 × 3 per second added) |
| Veo 3.1 Lite, 4, 6, or 8 seconds, 720p | Depends on the plan | 5 on Ultra, 10 for everyone else |
An 8-second line at 720p sits in the band of up to 10 seconds. A 15-second line uses the higher rate from second 11. Editing is on the same row as generating, so a fix of the 15-second clip spends another 60 credits. A 360p draft is the cheap listen: 8 credits for 8 seconds, 20 for 15. Ten generations of the 15-second 720p clip are 10 × 60 = 600 credits on today’s table, before any edit. That is arithmetic, not a promise of how many tries you will need.
Extending a clip in Flow is also Omni Flash 1.2 and needs Plus, Pro, or Ultra: 3 credits per second added at 720p, for 3 to 10 seconds. The API describes new speech on that extension when the starting video was generated by the model, not when you upload a clip of a person already talking. Veo 3.1 Lite stays as a 4, 6, or 8 second clip at 720p. The models page says Lite does not edit video-to-video and does not extend a clip, so a dialogue fix that needs an edit stays on Omni Flash 1.2.
The longer credit table, including the 5 December retirement of older models, is in the Omni Flash 1.2 credit guide. Unlimited Credit does not make a clip free. It means the shop refills the Flow pool with no cap on how many times it refills, from 10:00 to 21:00 Vietnam time (GMT+7), which is 0:00 to 11:00 in Brasília, with 2 to 3 hours between refills. Tell the bot when the pool hits zero. Outside that window the refill waits. This is not an instant, 24-hour service.
Which Veo 3 · Unlimited Credit plan fits
ChugAx is an independent shop in Vietnam. It sells access to Google AI Ultra 20x plans. It is not Google, and it is not a Google partner. Omni Flash still spends credits on every speaking take and every edit. The refill is what removes the cap.
| Plan | Who shares the pool | Price for the month |
|---|---|---|
| Veo 3 · Unlimited Credit · 1 member | The shop manages the Family. Up to 5 customers in that group share one pool. | 8,000,000 VND (about US$ 310.1, about R$ 1.538,46) |
| Full Family · Unlimited Credit | You manage the Family. You and 4 people you invite share the pool. | 40,000,000 VND (about US$ 1,550.4, about R$ 7.692,31) |
Prices are set in VND for the whole month. Reais use 5,200 VND per real and are a reference only. The amount you pay is the one the bot confirms. If you work alone and can accept a pool shared with other customers, the 1-member plan is the smaller bill. If the credits must stay with people you invite, choose Full Family. Ask the bot about the support tool that helps you use a shared pool.
If the 1-member plan stops during the month, the shop restores access or arranges a replacement account; if it cannot, it refunds the unused time. On Full Family the shop restores the plan when the problem comes from its own account; if it cannot, it refunds the unused time. Problems with your own Gmail are not covered. The three Full Family exclusions are in the warranty guide.
How to order from Brazil, and what still limits you
Orders go through the Telegram bot @chugax_gl_bot. The website does not take payment. There is no Pix key: Pix only moves money inside Brazil.
- Tell the bot which plan you want. It checks that your Gmail is eligible and states the amount.
- Pay only after that confirmation. Buy USDT with Pix and send it on the network the bot names (TRC20 or BEP20), or add reais to Wise with Pix and send VND to the MB Bank account the bot gives.
- The shop activates the plan 1 to 3 days after it has the details and confirms payment. That wait is for a new order, not for a later refill.
The steps in more detail are in paying from Brazil. Never send a password, an OTP, or a recovery code.
- Google has not evaluated Portuguese. A clear mouth in the English demo is not a Brazilian accent.
- Some voice options are regional. This page does not claim every Omni voice control is open in Brazil.
- Google and Flow limits still sit on top of the credit count, and the help page says the rates can change.
- Refills run 0:00 to 11:00 Brasília time, 2 to 3 hours apart. They are not instant and not 24 hours.