Flyne AI Launches Happy Horse 1.0: The Open-Source AI Video Model Ranked #1 in Blind Global Tests
Flyne AI today announced the public availability of Happy Horse 1.0, a new open-source video generation model that has achieved the top ranking on the Artificial Analysis Video Arena leaderboard in blind human-preference tests. The model is now accessible through the Flyne AI platform for creators, researchers, and production teams worldwide.
Happy Horse 1.0 has reached #1 globally in Text-to-Video without audio (Elo 1333) and #1 in Image-to-Video without audio (Elo 1392), surpassing closed-source competitors in head-to-head comparisons where outputs are evaluated anonymously. It also ranks #2 in both categories when audio is included.
What Makes Happy Horse 1.0 Different
Happy Horse 1.0 is a 15-billion-parameter model that generates video and synchronized audio together in a single pass, rather than combining separate systems for visual generation and sound. This unified approach produces more natural results where dialogue, motion, and ambient sound align coherently from the first frame.
The model supports native multilingual lip-sync across seven languages: English, Mandarin, Cantonese, Japanese, Korean, German, and French. Output quality reaches cinematic 1080p resolution for clips lasting 5 to 8 seconds in standard aspect ratios suitable for social media, advertising, and short-form content.
For developers and teams that need deployment flexibility, Happy Horse 1.0 is fully open source with commercial-use rights. Base model weights, distilled checkpoints for faster inference, a super-resolution module, and complete inference code are publicly available for self-hosting and fine-tuning.
Architecture and Performance
The model uses a 40-layer unified Transformer where text, image, video, and audio are processed as tokens in one sequence. The first and last four layers handle modality-specific encoding and decoding, while the middle 32 layers are shared across all modalities for efficiency.
Happy Horse 1.0 employs timestep-free denoising and 8-step DMD-2 distillation, which eliminates the need for classifier-free guidance and significantly accelerates generation. On a single NVIDIA H100 GPU, users can generate a 5-second 1080p clip in approximately 38 seconds.
Blind Rankings Reflect Real-World Quality
The Artificial Analysis Video Arena uses blind head-to-head voting where users see two anonymous outputs and choose their preference. Elo ratings are continuously recalculated from these votes, focusing on what humans actually prefer rather than self-reported technical specifications.
Happy Horse 1.0's #1 placement indicates it wins consistently on visual quality, prompt faithfulness, motion realism, and overall coherence—factors that matter most when creating content for real audiences.
Available Now on Flyne AI
Flyne AI has integrated Happy Horse 1.0 into its platform with dedicated workspaces for text-to-video and image-to-video generation. Users can experiment through the cloud interface without local GPU infrastructure while retaining the option to self-host for production workloads requiring privacy, custom fine-tuning, or cost control.
Happy Horse 1.0 demonstrates that open-source video generation can lead rather than follow closed systems when evaluated fairly. By making this model available on Flyne AI, we give creators and teams a practical path from testing to full production without being locked into proprietary APIs.
The architectural choices we made—unified self-attention with per-head gating—stabilize joint video and audio training enough to win in human preference tests. That stability is what enables 8-step inference, 1080p output, and multilingual lip-sync in a single pass.
Happy Horse 1.0 is a fully open-source AI video generation model released in early 2026. It jointly produces video and synchronized audio from text or image prompts with 1080p quality and multilingual lip-sync in seven languages. The model is available with commercial licensing, including base and distilled weights, super-resolution module, and inference code on GitHub.
Flyne AI is an AI image and video generation platform that brings multiple state-of-the-art models into a single accessible environment. It serves creators, marketers, educators, and researchers who need reliable high-quality visual output for social content, advertising, education, and creative storytelling.
For technical details, benchmark information, or interview requests:
Email: [email protected]
Website: https://flyne.ai/happy-horse-1-0/
Model Information: https://happyhorse.mobi
Leaderboard: Artificial Analysis Video Arena
