Silent AI video costs more: H3 always has sound, Kling 3 can drop it

If you want a video with no sound, the cheaper route on Sume is not a silent model: minimax-h3 always produces native stereo audio and costs $0.75 for 10 seconds at 768p, while the silent setting on kling-3 costs $1.40 for 10 seconds at 720p. The audio comes with the cheaper model, so mute it in the edit instead of paying more to avoid it.

All rates are provider list times 1.25 from Sume's Video Router docs, read on 2026-10-07. MiniMax's announcement describes H3's audio as 32 kHz stereo.

Which rows have a sound switch

Sume's docs say minimax-h3, minimax-h3-max and gemini-omni-flash-1.1 always produce audio, and generate_audio: false is rejected on them. kling-3 has an optional audio flag, with one price when audio is off and a higher one when it is on.

What 10 seconds costs

Kling's audio costs 50 percent more per second ($0.21 against $0.14). The difference on ten seconds is $0.70.

What to do for a silent deliverable

• Pick on look and price first, then remove the soundtrack in the final assembly.
• Use Sume's timeline or audio tools only if your workflow already has them; the video job itself will carry audio.
• Choose kling-3 with audio off only when you need Kling's motion or its 1:1 and 1080p options.
• Check the clip with sound on once, since a native track can contain speech you did not ask for.

Why a sound-free clip is not a cheaper clip

Prices on the video rows are per output second at a given resolution. Audio is a native part of the output on three of the four rows, so it is not an add-on that you can leave out to save money. Only Kling has a separate audio price, and even there the silent setting costs $0.14 a second, which is more than twice H3's $0.075 at 768p.

If silence is the goal, the savings come from picking the cheaper row and removing the track afterwards, not from a flag. If the clip will be played muted on a social feed, the audio may not matter at all and you can ignore it.

• Muted feed video: ignore the audio, pick on picture and price.
• Voice-over delivery: generate, then replace the track in assembly.
• Sound design: keep the native track and review it once.

A quick check for unwanted speech

Native audio can include speech that was not in your prompt. Listen to a draft with sound on before you deliver a clip that is meant to be quiet, and re-prompt with an instruction such as 'no speech, ambient sound only' if needed. Kling's silent setting is the one way to guarantee no track on a Sume row.

Choosing by what you ship

If the final product is a muted loop, the cheapest row at the smallest acceptable resolution wins, and the audio is irrelevant. If the product has voice, music or effects, a native track saves assembly work, even though it cannot be priced separately on most rows. Decide the audio plan first, since it narrows the list of rows more than price does, and then compare the remaining prices at the resolution you will ship.