Gemini Omni Flash tutorial

The better question behind Gemini Omni Flash

People search for Gemini Omni Flash expecting a magic button. The question that matters is how to chain Gemini 3.1 Flash TTS with reference audio inside Flixly so the output stays consistent across 60-second clips without extra credits.

We answer that directly. Start at the Text to Speech page, select Gemini 3.1 Flash TTS from the model list, upload a 15-second reference clip, and set temperature to 0.7.

Direct answer first

Gemini 3.1 Flash TTS inside Flixly produces 24 kHz mono WAV files at 1.2 credits per 10 seconds. Upload reference audio once, lock the voice ID, then generate multiple takes. The first generation takes 8 seconds on average.

Layers most guides skip

Reference audio length

Use exactly 12-18 seconds of clean speech. Shorter clips drop below 0.85 similarity score. Longer clips waste credits on the first upload.

Temperature and seed

Keep temperature at 0.7 for narration. Raise to 0.9 only for character dialogue. Save the seed number shown after each run so you can reproduce the same take later.

Export formats and sizes

Choose WAV 24 kHz for final delivery. MP3 128 kbps works for quick previews and cuts file size from 1.4 MB to 180 KB per minute.

Tradeoffs nobody mentions

Gemini 3.1 Flash TTS does not support singing or heavy accents without a second pass through Voice Cloning. Kling 3.0 handles video sync better when you need lip movement, but it costs 3 credits per second versus 0.12 credits per second for TTS.

Step-by-step instructions

1. Go to the dashboard and open the Text to Speech tool. Select Gemini 3.1 Flash TTS from the dropdown.
2. Paste your script into the text box. Keep sentences under 25 words for best pacing.
3. Click the reference audio upload button and choose a 15-second WAV file recorded at 48 kHz.
4. Set temperature to 0.7 and speed to 1.0. Note the voice ID that appears.
5. Hit generate. Review the first 10 seconds before queuing the rest.
6. If similarity is below 0.85, adjust temperature down to 0.6 and regenerate only that segment.
7. Export as WAV, then run the file through Auto Captions if you plan to post to video platforms.
8. Save the voice ID in your project notes so the next session loads the same profile instantly.

Credit cost example

A 45-second narration costs 5.4 credits. A full 10-minute podcast chapter costs 72 credits. Keep a buffer of 100 credits when testing new reference files.

FAQ

What sample rate should reference audio use with Gemini 3.1 Flash TTS? Record reference clips at 48 kHz mono. Flixly downsamples to 24 kHz internally, so higher source rates reduce artifacts.

How many seconds of reference audio give stable results? Twelve to eighteen seconds of clean speech produce similarity scores above 0.9. Anything under eight seconds fails the consistency check.

Can I reuse the same voice ID across different projects? Yes. Copy the voice ID string from the first generation and paste it into any new TTS session. No extra cloning credits apply.

Does Gemini 3.1 Flash TTS work with Lip Sync Video? It exports clean audio tracks that import directly into the lip sync tool. Sync accuracy stays within 40 ms on 1080p footage.

What happens if temperature exceeds 0.9? Output becomes inconsistent between takes. The same sentence may shift pitch by 30-40 Hz on repeated runs.

Is there a limit on daily generations for Gemini 3.1 Flash TTS? No hard limit exists, but accounts start with 500 credits. High-volume users buy packs at the [pricing](/ #pricing) section.

One decision rule

Lock the voice ID after the first successful reference upload, then reuse it inside the Text to Speech tool for every subsequent clip.