HappyHorse 1.1 para creators and developers
HappyHorse 1.1 — Alibaba's top-ranked video model, in your browser.
HappyHorse 1.1 launched (April 7, 2026) as a top-ranked frontier model on the Artificial Analysis Video Arena, going head-to-head with Veo 3.1, Sora 2, Kling, and every other frontier model on both text-to-video and image-to-video. Built by Alibaba's ATH AI Innovation Unit, led by Zhang Di (former Kuaishou VP, original Kling tech lead). On hilens, HappyHorse generations run on the same credit balance as every other model in Studio — $0.005 per credit at the Plus pack rate, no monthly subscription, credits never expire.
What makes HappyHorse a top-ranked model
HappyHorse uses a 40-layer unified self-attention transformer that generates video and audio jointly in a single forward pass — no separate cross-attention modules. Both text-to-video and image-to-video work natively, each with optional synchronized audio. That single-pass joint generation is what scores it ahead of Veo 3.1 (which also ships native audio) and well ahead of Seedance 2.0 and Kling v3-omni (which generate silent video and require audio added in post).
The Artificial Analysis Video Arena benchmark rates models on prompt adherence, motion coherence, visual fidelity, and physical plausibility — HappyHorse scored at the top across all four axes on debut, ranking among the leaders on both text-to-video and image-to-video categories, and it continues to hold a top-ranked position.
The team behind it is unusual: ATH (AI Innovation Unit) inside Alibaba's Taotian Group, started in late 2025 when Zhang Di — the original Kling tech lead at Kuaishou — joined Alibaba to build Future Life Lab. The Kling-pedigree shows in HappyHorse's motion handling.
When to pick HappyHorse over Sora 2, Veo, Kling, or Seedance
HappyHorse is the model to reach for when fidelity matters and audio is part of the deliverable — brand promos, narrative shorts, hero shots, product demonstrations where the video has to look and sound right in the same generation.
vs Sora 2: HappyHorse outranks Sora 2 on the Arena AND doesn't require a $200/mo ChatGPT Pro subscription or the now-shuttered sora.com — Sora 2 access has been a multi-quarter access lottery for paying OpenAI users. On hilens, HappyHorse is pay-per-credit on first signup.
vs Veo 3.1: Both ship native audio. HappyHorse currently rates higher on motion coherence; Veo is stronger on photorealistic single-subject shots. Choose by clip — many hilens creators run both on the same prompt and pick the winner.
vs Seedance 2.0: Seedance owns the short-form vertical motion-graphic niche (TikTok, Reels, Shorts). HappyHorse owns the premium-fidelity end. They're complementary, not substitutes.
vs Kling v3-omni: Kling is the right pick for brand-reference work (consistent characters, repeated subjects). HappyHorse beats it on one-shot premium output but doesn't yet match Kling's reference-image control.
How HappyHorse works on hilens — Studio
HappyHorse runs on the same credit balance that powers every other model on hilens. A $5 Plus pack covers HappyHorse generations alongside Kling, Veo, Seedance, and image models like Banana 2 — no per-model wallet to fund.
In Studio, pick HappyHorse from the video model picker, write your prompt, choose duration and aspect ratio. The cost in credits is shown on the Generate button before you click — typical HappyHorse generations sit in the premium frontier bracket alongside Veo 3.1, slightly higher than Kling. Failed generations are automatically refunded — you only pay for successful output.
Because HappyHorse access is still capacity-constrained at the platform level, hilens may queue HappyHorse generations at peak times. The Generate button shows queue position and ETA before submission.
Preguntas frecuentes
Preguntas frecuentes
- Is HappyHorse really a top-ranked video model right now?
- Yes — it has been a top-ranked frontier model on the Artificial Analysis Video Arena since its public debut on April 7, 2026, placing among the leaders on both text-to-video and image-to-video category leaderboards. Built by Alibaba's ATH AI Innovation Unit, led by Zhang Di (former Kuaishou VP and original Kling lead).
- How is HappyHorse different from Sora 2 and Veo 3.1?
- Sora 2 access has been gated behind ChatGPT Pro ($200/mo) and OpenAI shut down sora.com in March 2026, so the practical path to Sora 2 is now third-party platforms. Veo 3.1 requires Google AI Ultra or a Vertex-AI account. HappyHorse on hilens is pay-per-credit on first signup — no other subscription required. On the Arena benchmark, HappyHorse outranks both.
- Does HappyHorse generate audio in the same pass as video?
- Yes. HappyHorse uses a unified architecture that generates video and audio jointly in a single forward pass — both text-to-video and image-to-video have audio-enabled variants. Veo 3.1 also ships native audio. Seedance 2.0 and Kling v3-omni do not — they generate silent video and require audio layered in post.
- How much does HappyHorse cost on hilens?
- Each generation costs credits based on duration and resolution. Credits are $0.005 each at the Plus pack rate. HappyHorse sits in the premium frontier bracket on hilens — comparable to Veo 3.1, slightly above Kling. The exact credit cost is shown on the Generate button before you click. Failed generations are automatically refunded.
- Do I need a subscription to use HappyHorse on hilens?
- No. hilens is pay-per-credit — buy a pack and credits draw down only on successful generations. A $5 Plus pack covers HappyHorse alongside every other model in Studio, with no monthly fee, no quota reset, and credits that never expire. Failed generations are automatically refunded.