HappyHorse 1.0 — Alibaba's Open-Source Video Model That Topped the Leaderboards

HappyHorse 1.0 surfaced in April 2026 from Future Life Lab, a team inside Alibaba's Taotian Group, and quickly took the top spot on the Artificial Analysis Video Arena — the independent leaderboard that ranks video models on blind human preference — in both text-to-video and image-to-video, ahead of closed commercial models from ByteDance and Kuaishou. Its distinguishing feature isn't just the ranking: it's a 15-billion-parameter unified transformer that generates video and its matching audio in one native pass, not video first with a dubbing step bolted on, released under an Apache 2.0 license that permits commercial use.

Generate uncensored video now

What actually makes HappyHorse different

Most video models — including every model Uncutly currently runs — generate silent visuals only; if a clip needs a soundtrack or dialogue, that's a separate step outside the video model entirely. HappyHorse's single-stream architecture generates audio and video together in the same pass, including multilingual lip-sync. Combined with an Apache 2.0 license, meaning the released weights can be run, modified, and used commercially by anyone with the infrastructure to host them, that's a materially different distribution model than the closed, API-only releases from ByteDance (Seedance), Kuaishou (Kling), or MiniMax (Hailuo).

Where Uncutly's video lineup actually stands today

Uncutly.AI runs five video model families today — Kling, Seedance V1.5 Pro, Wan, Vidu, and Hailuo — and HappyHorse is not one of them. All five generate silent video; none does native joint audio generation. HappyHorse's open license makes it a plausible future candidate for adult platforms specifically, since an open-weight model can in principle be self-hosted and governed under a platform's own content policy rather than the model creator's, unlike a closed vendor API. That's a real, structural difference. It is not, however, a current feature of Uncutly or any integration we've shipped.

The gap between 'open' and 'available'

As of its debut, reports on whether HappyHorse's weights were actually downloadable were mixed — some coverage described the GitHub and Hugging Face pages as 'coming soon' rather than live. An Apache 2.0 announcement is a licensing intent, not proof that a usable checkpoint is sitting in a public repo today. Worth checking the primary source before assuming the weights are available right now.

What we're watching

If HappyHorse's weights become reliably available and its benchmark lead holds up under independent testing rather than just its own debut numbers, it's a genuinely interesting candidate for a self-hosted, audio-native video pipeline — including for adult content, given the open license. Nothing here is integrated into Uncutly yet.

FAQ

Can I generate video with HappyHorse on Uncutly.AI?

No — HappyHorse is not integrated into Uncutly.AI. Uncutly's current video lineup is Kling, Seedance V1.5 Pro, Wan, Vidu, and Hailuo.

Does HappyHorse generate audio?

That's its core claim — a single 15B transformer generating video and synchronized audio, including multilingual lip-sync, in one native pass, rather than video plus a separate dubbing step.

Is HappyHorse actually open source?

It was announced under an Apache 2.0 license, which permits commercial use. Some early reporting described the public weight release as 'coming soon' rather than immediately available — worth verifying current status at the source before assuming the checkpoint is downloadable today.