Multimodal Transformer 2

Open-source 15B parameter AI model for joint video and synchronized audio generation. Happy Horse 1.0 is an advanced open-source AI video generation model designed to produce high-quality 1080p videos with synchronized audio and multilingual lip-sync capabilities. Developed by the Happy Horse team in 2026, it utilizes a 15-billion-parameter unified Transformer architecture to jointly generate video frames and corresponding sound from text or image prompts. The model is fully open-source, including weights and inference code, and is designed for self-hosting with full commercial-use rights. It features a unique 40-layer self-attention network that ensures stable training and cinematic output, making it a powerful tool for professional video production and localized content creation.

Open-source SOTA AI video generator with native audio-video sync and fast 1080p output. Happy Horse 1.0 is a state-of-the-art open-source AI video generator that utilizes a 15B-parameter unified Transformer architecture to produce high-quality videos from text or images. It is distinguished by its native joint audio-video synthesis, which generates perfectly synchronized dialogue, ambient sounds, and foley effects in a single pass. The model features DMD-2 distillation, allowing for incredibly fast inference—generating 1080p videos in approximately 38 seconds. It also offers industry-leading lip-sync support for seven languages, making it a powerful tool for global content creation, marketing, and filmmaking.