Bad news on Happy Horse from twitter

HappyHorse-1.0 is a pseudonymous AI video generation model that appeared on April 7, 2026, topping the Artificial Analysis Video Arena leaderboard in both text-to-video and image-to-video (no audio) categories, surpassing established models like Seedance 2.0 and Kling 3.0, despite having no publicly known development team or official launch. Although the model's website (happyhorses.io) claims it is "fully open source" with a 15B-parameter unified self-attention Transformer architecture supporting joint audio-video generation, no weights, inference code, or GitHub repository have been made publicly available as of early April 2026 — with all linked resources returning "coming soon" pages or 404 errors. Community speculation ties the model to Sand.ai's open-source daVinci-MagiHuman project, and the Reddit/r/StableDiffusion thread reflects broader community concern that the "bad news" is the gap between HappyHorse's impressive benchmark performance and the lack of any accessible, locally-runnable release.

Related

• HappyHorse is from Alibaba ATH, not Grok / Veo 3.2 / Wan 2.7 / Seedance 2
• Live AI video is doing too much lifting as a term. Here's a breakdown of what people actually mean.
• Got early access to a real-time interactive video model, here's what I found
• LTX 2.3 Lip Sync Music Clip -- Drake - Toosie Slide
• LTX 2.3 - Image + Audio + Video ControlNet (IC-LoRA) to Video

Source: local-ai