Happy Horse: The Seedance Killer? Plus BIG 4k AI Video News!

Alibaba quietly launched HappyHorse-1.0 and it shot to #1 on the Artificial Analysis video leaderboard, with a lot of people calling it a Seedance killer. Short answer: it's not — at least not yet. But the team behind Happy Horse is led by Zhang Di, the architect of Kling 1.0 and 2.0, who left Kuaishou last fall, joined Alibaba in November, and shipped this in roughly five months. So yeah, this is a pony worth keeping an eye on. Also in this one: Topaz drops Starlight Precise 2.5 (and surprise-launched a new creative model mid-shoot), Kling goes native 4K (real 3840×2160, no sneaky upscaler), and Netflix's Eyeline Labs releases Vista4D — an open-source 4D reshooting framework that lets you retarget the camera on existing footage. Open source. From Netflix. 2026 is weird. LINKS: Happy Horse: https://www.happyhorse.com TOPAZ: https://topazlabs.com/ref/2518/ Kling: https://klingaiaffiliate.pxf.io/qz3955 Vista 4D: https://github.com/Eyeline-Labs/Vista4D Signup for my Newsletter: https://theoreticallymedia.beehiiv.com/ Chapters: 00:00 Big AI video updates 00:55 Happy Horse from Alibaba Arrives 01:10 Leaderboard hype and early buzz 01:30 Who made Happy Horse? 01:56 Happy Horse testing begins 02:18 First text-to-video results 02:38 Prompting discovery: keep it short 02:59 Prompt format and token spam 03:23 Why brevity matters 03:43 Image-to-video test: Twin Peaks diner 04:08 Lip sync and voice issues 04:40 Prompt adherence wins 04:57 Kung-fu/action limits 05:11 Can Happy Horse handle action? 05:39 Current limitations and missing features 05:58 1080p, 15-second generations, and batch output 06:08 Reference / Omni mode 06:38 Reference mode problems 06:57 Prompt troubleshooting 07:17 Reference mode working well 07:57 Is Happy Horse a Seedance killer? 08:18 Topaz update begins 08:49 Starlight Precise 2.5 test 09:12 4K upscale detail check 09:23 Cleaning without changing faces 09:44 Skin texture test 09:59 Another Seedance upscale test 10:15 Precise mode vs creative detail 10:36 The unhinged Bruce Lee Terminator test 11:05 Starlight Precise 2.5 is available 11:17 Astra Creative 2 drops 11:36 Kling native 4K generation 11:55 Robotech/Macross 4K test 12:18 Why native 4K matters 12:34 Stock and travel footage use case 13:03 When to actually use 4K 13:18 Kling 4K verdict 13:35 Every AI video model has strengths 13:56 Netflix Vista4D 14:13 How Vista4D works 14:25 Compared with Google Flow and Veo 3 experiments 14:40 Camera control limits 15:01 Video outpainting and Wan 2.1 15:11 Where to try Vista4D 15:22 Wrap-up