Kling VIDEO 3.0

4K here means 4K

Kling opened native 4K generation on the 3.0 series on 23 April, and its own announcement is explicit that the output comes straight from the model rather than from an upscaler. Anyone who has worked with upscalers knows the difference: they shift faces, and the character drifts between shots. Generating at the target size does not have that problem. In our video table this is the only model that takes the resolution column outright.

Fifteen seconds, with a free length

Kling own guide sets duration anywhere from three to fifteen seconds. Fifteen is more than Veo eight and less than Seedance thirty, and that middle is exactly where it belongs: enough for an advertising shot or a short sequence, not enough for a whole story.

The audio, and the language it does not have

Audio in this series is native: speech, lip movement and scene sound are generated with the picture, and in a multi-character scene you can point at which character is speaking. The guide counts the speech languages and there are five: Chinese, English, Japanese, Korean and Spanish, plus dialects and accents. Persian is not on that list.

For a Persian-speaking audience that is the most important sentence on this page: this model will not produce a Persian spoken video, and the voice has to be added separately. The same limit applies to the other three models in the table. Not one of them lists Persian among its speech languages.

References: seven images, or four if you supply a video

The reference number is published in the Omni mode guide and it is precise: up to seven images or elements with no video, dropping to four in total once you add a video. An element is several images of the same subject from different angles, up to four of them, combined into one subject, and you can bind a voice sample to it.

One constraint from the same guide carries weight in our table: Omni, the reference-driven path, outputs 1080p and 720p only. So 4K and heavy referencing do not happen in the same generation.