Veo 3.1 audio is always on at Google; Sume's Veo rows let you mute

At Google, Veo 3.1 audio is always on, according to the Gemini API Veo page (read 2026-10-10). Sume's Veo rows take an optional generate_audio flag that defaults to true, and setting it to false lowers the billed rate by roughly half on the full model.

The catch is availability. Sume's Veo rows appear in a workspace only when GET /v1/videos/models returns the id, so check your catalog before you rely on this.

What each side says about audio

Google's page (read 2026-10-10) states that audio is always on for Veo 3.1 and that every output carries a SynthID watermark. It lists no switch to mute the track. Runway's pricing page (read 2026-10-10) shows Veo 3.1 at 40 credits per second with audio and 20 without, so Runway does expose a muted price.

On Sume, the request builder for Veo in the API code treats generate_audio as an optional boolean that defaults to true. The catalog description says the rows are text-to-video with optional synchronized audio.

The muted rate on Sume

Sume bills Google's list price times 1.25 at submit. The pricing code lists Veo at 720p per second as $0.40 with audio and $0.20 without for veo-3.1, $0.10 and $0.08 for veo-3.1-fast, and $0.05 and $0.03 for veo-3.1-lite. The table shows the billed result for an 8-second clip.

The fast row saves the least in relative terms: muting takes it from $1.00 to $0.80, a 20 percent cut, while the full model drops by half.

A batch example

Say you render 50 six-second vertical teasers and add your own music. At 720p, billed, a 6-second clip with audio costs $3.00 on veo-3.1 and $0.75 on veo-3.1-fast; muted it costs $1.50 and $0.60. Across 50 clips that is $150.00 versus $75.00 on the full model, and $37.50 versus $30.00 on fast. The saving is real on the full model and modest on fast.

Arithmetic here is per-second rate times 6 seconds times 50 clips, with the rates from the table above.

When muting is the right call

Mute when you will replace the soundtrack anyway: a voiceover, licensed music, or a dubbing pass. The generated track is then dead weight on the bill, and on the full model it is half of it. Keep audio on when the clip is final and you want the ambient sound that Veo synchronises to the picture.

Remember that you cannot mute at Google, so a workflow that depends on a silent Veo file has to strip the audio after download there. On Sume that step is not needed when the row is listed, because the flag is part of the request.

• Send generate_audio: false to get the lower rate on a Veo row.
• Read generate_audio in the catalog entry to confirm a row can produce audio at all.
• Keep the Idempotency-Key header on submits so a retry does not bill twice.

Check the catalog first

Sume documents GET /v1/videos/models as the place to discover models and their prices, and the Video Generation page shows the response fields. If no veo-3.1 id is returned, your workspace does not carry the row, and the prices above do not apply. The earlier post linked below covers what Sume lists in that case, and the Errors and rate limits page explains the 402 you get when the balance cannot cover a reservation.