ElevenLabs TTS on this canvas is 13 credits per 1,000 characters, which makes narration one of the cheapest things you can generate. A 90-second script costs about 18 credits. A full-length book costs about $65.
The arithmetic is simple. The craft is punctuation and a proofread.
TL;DR
- 13 credits per 1,000 characters, about $0.13
- 90-second script (~1,400 chars): 18 credits. 3-minute script: about 36
- Punctuation is your pacing control, because synthetic delivery follows it literally
- Read it back once. Names, jargon and numbers are where it fails, and those are the words that matter
- Library voices, not cloning. This is TTS, not a voice cloning platform
The arithmetic
Canon 8frame prices, a credit is $0.01 at pack rate:
| Script | Characters | Credits | Approx |
|---|---|---|---|
| 15-second social clip | ~230 | 3 | $0.03 |
| 30-second ad | ~450 | 6 | $0.06 |
| 90-second explainer | ~1,400 | 18 | $0.18 |
| 3-minute video | ~2,800 | 36 | $0.36 |
| 10-minute video | ~9,000 | 117 | $1.17 |
| 90,000-word book | ~500,000 | 6,500 | $65 |
Speaking pace is roughly 150 words a minute, and a word averages about 6 characters with spaces. So characters divided by 900 gives you approximate minutes.
Punctuation is the interface
Synthetic delivery follows punctuation more literally than a person does, which is a control rather than a limitation once you know it.
Commas are short breaths. Use them where you want the voice to pause, even where a grammar checker would not.
Full stops are longer breaths. A long sentence without any will be read without air and sounds rushed.
Paragraph breaks are the longest pauses. Use them between beats.
Em dashes and semicolons behave inconsistently. Rewrite around them.
Write numbers as words where pronunciation matters. "$19" is ambiguous; "nineteen dollars" is not. "1985" can come out as "one thousand nine hundred eighty five".
The proofread that matters
Listen to the whole track once. Not a spot check.
What fails predictably, and it is always the important words:
- Proper nouns. Your product name, your company name, people's names.
- Numbers and units. Prices, dates, measurements.
- Domain jargon. Any term the model has not heard often.
- Acronyms. Read as letters or as words, inconsistently.
A 90-second track takes 90 seconds to check. Shipping a video where your product name is mispronounced for the whole runtime is a self-inflicted wound.
The fix for a stubborn word: spell it phonetically in the source text, or use the tool's pronunciation dictionary if you have one.
When to use your own voice instead
Synthetic narration is genuinely good for informational delivery, which is where the register is flat anyway. It is weak where the voice has to carry conviction.
Use synthetic for: documentation, training, tutorials, explainers, product walkthroughs, localised versions of your own material.
Use your own voice for: anything about your judgement, your company's position, a founder message, an apology, or a video where the audience is deciding whether to trust you. The audience is responding to presence, and presence is the thing TTS removes.
Scope, stated plainly
This is text to speech with library voices. It is not a voice cloning platform. If you want narration in your own voice, that is a dedicated service, and the generated audio can then be used in your work here.
And the rule that applies wherever cloning happens: your own voice is legitimate, anyone else's needs documented consent for the specific use, and never a public figure.
The full narrated-video cost
| Element | Route | Credits |
|---|---|---|
| Narration, 1,400 characters | ElevenLabs TTS | 18 |
| Twelve illustrative stills | Seedream 5 at 6 | 72 |
| Two b-roll plates, 5s | Wan 3.0 at 480p | 48 |
| Music bed, 30s looped | Lyria 2 | 13 |
| Assembly | Merge Videos | 10 |
| Total for 90 seconds | 161, about $1.61 |
FAQ
How much does AI narration cost? 13 credits per 1,000 characters. A 90-second script is about 18 credits.
How do I control the pacing? Punctuation. Commas and full stops are breaths, and synthetic delivery follows them literally.
Does 8frame clone voices? No. Library voices only.
Do I need to check the output? Yes, all of it. Names, numbers and jargon are where it fails.
Punctuate for breath, listen to all of it once. The 8frame canvas is free and unlimited, and generation is paid from $19/month.