Kling AI releases version 4.0 video generator with stereo audio

By Splitfeed EditorialLast updated Oct 19, 2026, 17:40 UTC

Kuaishou released Kling AI version 4.0 on October 19, 2025, bringing several concrete upgrades to its video generation model. The new version supports clips up to 30 seconds in length, accepts up to ten keyframe images as input, and extends prompt comprehension to 8,000 tokens. It also introduces stereo audio output and multilingual lip sync, features that move the tool closer to a production-ready pipeline for creators working across languages. A lighter Flash variant was released alongside the main model, suggesting Kuaishou is trying to serve users with different compute or cost constraints.

The more consequential detail in this release is not what Kling 4.0 added but how it compares to its rivals. In benchmark comparisons against Wan 3.0 and Seedance 2.5, the new model performs at a comparable level but does not clearly outperform either. That matters because the AI video space has been defined by rapid, leapfrogging improvements over the past two years. If a major version update from a leading model lands roughly level with the competition rather than ahead of it, it raises a genuine question about whether the pace of capability gains in this category is beginning to plateau. That question is not settled by a single release, but it is the context that makes the benchmark result worth noting beyond the feature list.

None of the three camps have published reactions to this specific release. Pro-AI commentators would typically treat a feature-rich update like Kling 4.0 as further evidence of the practical maturity of generative video tools, pointing to stereo audio and multilingual lip sync as markers of real-world utility. Anti-AI voices would likely focus on the implications for creative workers, particularly those in dubbing, localization, and short-form video production, arguing that multilingual lip sync accelerates displacement in those roles. Middle Ground observers would probably acknowledge the incremental gains while treating the competitive stalemate with Wan 3.0 and Seedance 2.5 as a reason to wait before drawing conclusions about which platform deserves adoption.

The argument about whether AI video progress is leveling off will be tested by the next round of model releases from Kuaishou, the teams behind Wan and Seedance, and others active in the space. Independent benchmark results published after wider access to all three models would offer a more reliable basis for comparison than any single company's release claims.