Best way seriesNovember 16, 2026
The best way to make a product demo video with AI
Have your agent record one silent pass through your product, then call vaaya/produce_autodemo once. A six-section demo costs at most $2.44 at listed prices.
Have your agent record one silent pass through your product, then make a single call to vaaya/produce_autodemo. Vaaya watches the recording, writes and voices the narration, edits the footage with cuts and zooms, and adds music and captions before it returns one finished video.
Vaaya is a wallet for AI agents. You recharge it with a prepaid pack and your agent pays per call from that balance. There is no subscription and no separate account or API key for each service a demo draws on.
What "best" means for a demo video
- It shows the real product. The footage is a recording of your live app. Vaaya's routing guidance is firm that generated video or a simulated screen is no substitute.
- Little work by hand. You do not write a script or sit in an editor.
- A finished file comes back. One video with narration, zooms and captions already in it.
- A cost you can bound before you start. Every step has a listed price, so the ceiling is arithmetic.
The options in the catalog
Prices are as listed on 16 November 2026. The catalog has current prices.
| Tool | What it returns | Price per call | When to pick it |
|---|---|---|---|
vaaya/produce_autodemo |
A job_id, then the finished narrated and captioned video |
0¢ for the call. The steps inside it are charged: at most $2.02, plus 7¢ per narrated section | Any product demo. It is the one demo path consult offers |
files/upload |
A file_id and an upload URL for the recording |
1¢ | Before the call above, to store the recording |
vaaya/result |
The job status, then the video URL | 0¢ | To poll the job. Do not call vaaya/produce_autodemo again to check on it |
fal/generate with elevenlabs--tts--turbo-v2-5 |
A voiceover audio file | 5¢ for each started 1,000 characters | You want narration only and will edit the video yourself |
deepgram/speak |
A hosted MP3 | 1¢ for each started 250 characters, 2¢ minimum | Draft or utility narration, up to 2,000 characters per call |
For narration alone, the voiceover guide walks through one script. Generated clips are a different job, covered in the video model comparison.
The default path: one recording, one call
- Capture. The agent opens your product in a local browser with Playwright and records the screen, using
apertureon macOS orffmpegon Windows. Both ship with Vaaya's MCP package. It tours the feature for 30 to 160 seconds and logs each click, highlight and typing target with a timestamp and a screen position. If the product needs a login, you sign in yourself in that window. - Describe. The agent writes four fields:
whatItDoes,builderIntent,companyanduseCases. The recording shows what happens. These fields supply the reason. - Hand off. The agent stores the recording with
files/upload, then callsvaaya/produce_autodemoonce. - Produce. Vaaya plans the edit from the recording, trims and speeds up slow stretches, zooms to the logged positions, voices the narration, adds a music bed, renders, and burns in captions.
- Deliver. The agent polls with the
job_iduntil the video URL comes back.
Step outside this path when you need to control the script word for word or want a shape other than 16:9. Then use the narration tools in the table and your own editor.
A worked example: what one run costs
vaaya/produce_autodemo is listed at 0¢ because the steps inside it are charged one by one to the same balance:
- Narration: 5¢ for each started 1,000 characters, per narrated section
- Storing each generated audio file: 1¢
- Music bed: 15¢
- Staging each file for the edit (the recording, each narration clip, the music): 1¢
- Creating the project, validating the edit and writing it: 1¢ each
- Render: its actual price, never more than $1
- Captions: 80¢
The number of narrated sections is set when Vaaya plans the edit, so you learn it after the run. Take a 75-second recording whose plan narrates six sections, each under 1,000 characters. This is arithmetic at listed prices on 16 November 2026. It reports no measured results.
files/upload for the recording 1¢
Narration, 6 x 5¢ 30¢
Store 6 narration clips and the music 7¢
Music bed 15¢
Stage 8 files for the edit 8¢
Project, validation, edit 3¢
Render up to 100¢
Captions 80¢
Total up to 244¢ ($2.44)
The general rule: at most $2.02 for the fixed steps, plus 7¢ for each narrated section.
The calls the agent makes are files/upload with filename, size_bytes and mime, then this one, then vaaya/result with the job_id:
{
"service": "vaaya",
"action": "produce_autodemo",
"params": {
"recording": "<file_id from files/upload>",
"feature": {
"whatItDoes": "Exports any invoice list to CSV or PDF in two clicks",
"builderIntent": "Finance teams were copying rows by hand at month end",
"company": "Acme makes invoicing software for small agencies",
"useCases": "Month-end close, and handing data to an accountant"
},
"clicks": [
{ "t": 4.2, "x": 0.62, "y": 0.31, "kind": "click", "intent": "the Export button" }
],
"recordingDurationSec": 75
},
"max_cost_cents": 0
}
How to run it
Connect an agent that runs on your own Mac or Windows machine, because it has to record your screen. Then ask in plain English:
Use Vaaya to make a narrated demo video of the invoice export in my app at app.example.com. Record it yourself in a local browser, and I will sign in when it opens. Show the filter, the export menu and the downloaded file. Before you start, tell me the most this run can cost. Afterwards, list what each step was charged.
The agent can ask Vaaya's consult tool to plan the calls. Every use call carries max_cost_cents. On this job that ceiling is compared with the call's own listed price, which is 0¢ for vaaya/produce_autodemo, so it does not cap the steps inside the job. Each step is checked against your balance when it runs. Work out the ceiling with the arithmetic above and confirm the balance covers it before the call.
What to check, and where this falls short
- Watch the whole video. A model writes the narration from the recording and your four fields. Check product names, numbers and claims.
- Zooms. They land on the positions the agent logged. Without that log, Vaaya infers the focus from the recording, with lower precision.
- One format. The output is 16:9 at 30 frames per second in full HD, with no presenter or avatar. Captions come in one style and are set to US English.
- No edit pass. The call has no parameter for changing the script or the cut. You steer with the recording, the four fields,
targetDurationSecandvoice(defaultLiam). A second run is charged again. - Capture limits. The recording should run 30 to 160 seconds and fit the 100 MB file limit. Linux capture is not supported yet.
- Charges are per step. A failed step is not charged, and neither is a failed render. Steps that finished before a failure stay charged. If the music or caption step fails, the video ships without it. The
charged_centsin the poll result does not add up the steps, so read the run from Transactions in your account.
Why this job suits a pay-per-call wallet
A demo is occasional work: you make one when a feature ships, then none for weeks. One run also draws on separate services for voice, music, editing and captions, and you hold an account with none of them. You pay for the steps of the runs you make, each listed on its own line. If a charge looks wrong, you can raise a dispute from Transactions in your account, and a person at Vaaya reviews it. Packs are on the pricing page.
Questions
How much does an AI product demo video cost?
On Vaaya the vaaya/produce_autodemo call is listed at 0¢ and the steps inside it are charged one by one. At prices listed on 16 November 2026 the fixed steps total at most $2.02 per run and each narrated section adds 7¢, so a demo with six narrated sections costs at most $2.44.
Do I have to record the screen myself?
No. Your agent drives the product in a local browser and records the screen on macOS or Windows, for 30 to 160 seconds. If the product needs a login, you sign in yourself in that browser window. Linux capture is not supported yet.
Can I edit the script or the cut afterwards?
The call has no parameter for editing the script or the cut. You steer the result through the recording, the four-field feature description, targetDurationSec and voice. A second run is a new job and is charged again.
What does the finished video include?
A 16:9 video at 30 frames per second in full HD, with narration, zooms on the moments the agent interacted with, sped-up or trimmed slow stretches, a music bed and burned-in captions. There is no presenter or avatar.