Gemini TTS

MODEL COMPARISON / SAME SCRIPT, TWO MODELS

Gemini 3.8 Flash vs Flash-Lite TTS

Choose a model for the recording you need. Compare Flash and Flash-Lite with controlled scripts, the same available voices, and credit estimates from this application. Use a short passage to judge clarity and delivery before committing to a longer recording.

Compare the test scripts

One narrator · the same practical explanation

Read the complete test script and direction
A first draft makes an idea easier to discuss. Choose one person who needs the result, then write a short example of what they would do. Read it aloud and remove any sentence that repeats the point. Share that version before adding more detail. The useful question is not whether the draft contains everything. It is whether someone can understand the next step and explain why it matters.

A clear, friendly narrator. Natural pace, neutral accent. Explain the idea without dramatic emphasis.

Voice: Kore · Neutral accent · Default style and pace

Flash · English · Generated for this site

Recorded 2026-10-08

Try this script with Flash

Flash-Lite · English · Generated for this site

Recorded 2026-10-08

Try this script with Flash-Lite

Two speakers · the same question and reply

Read the complete test script and direction
Speaker 1: Why do you start with such a small example when the project is much bigger? Speaker 2: Because a small example gives us a clear question to test. We can ask someone to try one action and explain what was confusing. Speaker 1: But does that tell us enough about the final version? Speaker 2: It tells us about the next step. We still need to test the rest, but we can improve the explanation before building every feature. Speaker 1: So the first draft is a way to learn, rather than a promise that everything is finished? Speaker 2: Exactly. Keep the test focused, listen to the feedback, and choose the next change from what you actually observed.

A relaxed conversation. A curious host and a measured guest explain a practical decision clearly. Natural pace, neutral accents.

Voices: Kore / Fenrir · Neutral accent · Default style and pace

Flash · English · Generated for this site

Recorded 2026-10-08

Try this script with Flash

Flash-Lite · English · Generated for this site

Recorded 2026-10-08

Try this script with Flash-Lite

How we keep the test conditions the same

A comparison starts with a repeatable input. Use the same text, speaker voices and optional controls for Flash and Flash-Lite. Changing the story, language or voice at the same time makes it difficult to tell what caused the difference. The narration and dialogue scripts on this page are separate test cases. Within each case, hold the settings constant and review the recordings from beginning to end.

The existing welcome, Mandarin reading and dialogue examples elsewhere on the site demonstrate individual recordings. They use different text and should not be treated as a matched comparison between models. For your own test, include a representative name, number or phrase from the project. A short sample is useful for selecting a starting point, but it cannot establish performance across every voice, language or long-form passage.

Compare application credits for the same text

Both models use started blocks of 100 submitted characters in this application. Flash costs 1.2 credits per block and Flash-Lite costs 0.8. A 100-character script uses one block; adding one character starts another. A 500-character passage therefore estimates 6 credits with Flash or 4 with Flash-Lite. These are application credits, not a quotation for the Google API or a promise about the length of the resulting audio.

The table below uses the same pricing calculation as the studio. For dialogue, the combined submitted text contributes to the estimate, including supported vocal tags and the separator used by the request contract. Check the live estimate after every edit or model change. New accounts receive 20 starter credits valid for 30 days. Repeated takes are separate generation requests, so keep the first tests focused and download any version you want to retain.

Application credit estimates · per request
Model100 characters1,000 characters
Flash1.2 credits12.0 credits
Flash-Lite0.8 credits8.0 credits

When to start with Gemini 3.8 Flash TTS

Start with Flash when the performance is a major part of the work: a story opening, character exchange, or narration with a change in tone. Choose a voice, write a concise description, and give the script a clear structure. More instructions are not always more useful. Begin with one delivery goal and listen for whether the result conveys it before combining several moods or adding cues throughout the text.

The Flash model page includes the full studio and editable examples. Use those to prepare a first take, then return to your comparison with the same passage. Listen for how the reading moves between sentences and whether the two roles remain easy to distinguish. These are practical listening questions, not an objective score or a guarantee that Flash will be the preferred model for every script and listener.

When to try Gemini 3.8 Flash-Lite TTS

Try Flash-Lite when the lower application credit rate helps you review more script variations. A straightforward explanation or read-aloud draft is a useful first test. The studio still exposes voices, accent, style, pace, voice descriptions and scene direction. Select the controls you actually need rather than assuming that a lower credit rate requires a completely different writing process. You can keep the script and change only the model.

Flash-Lite generation time can vary with request conditions; this page does not promise it will always finish sooner. Compare the clarity and delivery of the completed recording for your use case. If a name or transition needs work, edit the wording before deciding that switching models is the only solution. The Lite model page provides a dedicated starting point and current credit examples without changing your saved recordings.

The same studio, voices and WAV workflow

Both models here support one narrator or up to two speakers, accept a combined script within 5,000 characters, and return completed audio with a WAV download. Voice previews are existing neutral recordings; listening to them does not submit a task. Applying a script template also does not generate audio. Generation requires signing in and enough credits for the selected model and submitted text.

A model switch configures the next request. It does not change an existing take, convert the result, or automatically produce a second recording. Keep your approved file and settings if you plan to build a series. For a two-host programme, use the podcast page to organise the script; for general role assignment, use the two-speaker guide. Subsequent editing, separate speaker tracks and automatic background music are outside the current studio.

Use a simple listening checklist

Listen once for understanding, then again for delivery. Is the meaning clear without reading along? Are names and numbers understandable? Does a pause help the next idea land? Can you distinguish the speakers without looking at their labels? Use ordinary speakers as well as headphones if that reflects where the audience will listen. Record your preferred settings alongside the script so the next test has a clear starting point.

Keep the scope of your conclusion small. One successful short test shows what happened with that passage and those settings, not every language or an hour-long programme. If the result is inconsistent, review another representative segment before making a production decision. You can also change one variable, such as the voice description, while leaving the model unchanged. A useful comparison helps you choose a recording you can actually use.

Questions before you start

Which model is cheaper in this application?

Flash-Lite has the lower configured credit rate: 0.8 credits per started 100 characters, compared with 1.2 for Flash. The table and live studio estimate use the application pricing calculation.

Can I switch models without rewriting the script?

Yes. Select another model in the studio and review the new estimate. The switch prepares the next request; it does not regenerate an existing take.

Are the model languages identical?

Official language coverage differs by model. Check the current Google Gemini API documentation and test your intended text in this application; a listed language does not establish the quality of every voice.

Is this a speed or quality benchmark?

No. These scripts support a controlled listening comparison and a choice for your own project. They do not establish a universal ranking or guaranteed generation time.