Kling 3 audio on or off: 8-second clips cost 168 or 112 cents on Sume

kling-3 on Sume costs 112 cents for an 8-second clip with audio off and 168 cents with audio on, and the price is the same at 720p and 1080p. For 250 prompts that is $280.00 silent or $420.00 with sound.

Kling is the only one of the six candidate models where the audio is a price decision. After OpenAI removed the Videos API and the sora-2 models on 2026-09-24, a stored library with narration or ambient sound needs a plan for the audio track, and Kling's switch is worth knowing about.

Both prices at every length

The list price is $0.112 per second silent and $0.168 per second with audio, times 1.25 and rounded up to the cent. The audio premium is exactly 50 percent.

Why 1080p is free here

The grid above is identical for 720p and 1080p. If you route a library to kling-3 at all, ask for 1080p when the deliverable is full size: an 8-second 1080p clip is 112 cents, the same as at 720p. Wan 3.0 costs 100 cents at 720p and 200 at 1080p for the same 8 seconds.

That makes Kling the low-cost 1080p line for silent 8-second clips: 112 cents against 200 for Wan, 341 for seedance-2-mini and 545 for seedance-2-fast.

Library totals at 5, 8 and 12 seconds

Take a library of 250 prompts that are all the same length. At 5 seconds it costs $175.00 silent or $262.50 with audio. At 8 seconds it costs $280.00 or $420.00. At 12 seconds it costs $420.00 or $630.00. The premium grows with length, from $87.50 to $210.00 across those three lengths.

If only some clips need sound, split the library by need and send generate_audio: false on the rest. On a 250-prompt library where 100 prompts need audio, the bill is 100 x 168 + 150 x 112 = 33,600 cents, or $336.00, against $420.00 for audio on every clip.

Controlling the audio switch

Use the generate_audio field in the /v1/videos request body. The Video generation docs describe it as telling the model to generate audio or not, with the default set by the model's audio capability, so send it explicitly if the price matters.

Kling-3 does not take reference image, video or audio URLs, according to the Sume catalog. If your Sora prompts depended on a reference image, use first-frame frame_images instead, or choose a model that accepts references.

• Silent is the right choice when you add a voice-over or music in an editor later.
• Audio on adds exactly half again to the price of every clip.
• At 1,000 clips of 8 seconds the switch is worth $560.00.