New AI Video Model ‘Happy Horse’ Tops Leaderboards
A new artificial intelligence model for generating videos, named ‘Happy Horse’, has quickly climbed to the top of analysis leaderboards, even surpassing the well-regarded Dreamina Seance 2.0. This development marks a significant step forward in AI video creation, offering impressive quality and, notably, the promise of being open-source.
Happy Horse Emerges Victorious
The artificial analysis leaderboard, a common way to compare AI models, shows Happy Horse with a substantially higher ELO score than Seance 2.0. This isn’t a small margin; Happy Horse has nearly 100 more ELO points, a surprising achievement given the competitive field that includes models like Cling 3 and Sky Reels. While some suggest the high ranking might be due to people intentionally voting for the new model, early examples suggest Happy Horse is genuinely a powerful contender.
Developed by Alibaba, Happy Horse is described as being more in line with recent Cling models but potentially better. It also features native audio generation, which is a significant plus. The model appears to generate video at 720p resolution, offering great detail and smooth movement. Multi-shot sequences seem to be a particular strength, with examples showing believable backgrounds and realistic interactions, like the crunch of boots on ice.
While Happy Horse excels in many areas, it’s not without its quirks. Some users have noted that the audio can sometimes sound a bit dramatic or cartoony. In one example, voices sounded overly intense, like meteors were falling. Physics in action sequences, like a basketball game, are generally impressive, though sometimes the speed of movement can look more like slow-motion footage than intended. However, the detail in scenes with multiple people and the physics of a basketball net are highlighted as strong points. The model also shows promise in generating complex scenes, like a science-themed video with a telescope, though the accuracy of scientific equipment might not always be perfect. Notably, it avoids morphing issues and maintains believable human figures in its generations.
Open Source: The Big Advantage
What truly sets Happy Horse apart, beyond its performance, is its planned open-source release. This is a major contrast to models like Seance 2.0, which are closed-source and have only recently become available in the US, with limited API access. An open-source model means researchers and developers worldwide can access, study, and build upon it freely. This accessibility can accelerate innovation across the entire AI community.
The creator, identified as Alibaba, has made Happy Horse available for testing on an arena where users can rank generated videos. While you can’t input your own prompts in this testing environment, it allows for direct comparison. In these tests, Happy Horse has shown an ability to follow prompts closely, even in complex scenarios like seasonal changes or intricate character movements. While it can sometimes produce less-than-perfect results, particularly with rapid motion where some warping might occur, it often corrects itself quickly.
Seance 2.0: A Strong Competitor with Caveats
Dreamina Seance 2.0, despite being surpassed on the leaderboards, remains a powerful AI video model. Its recent availability in the US is a notable event, especially considering the strict safety filters implemented by its creators. These filters limit the generation of violence, gore, nudity, and even the uploading of real human faces. While these restrictions aim to ensure responsible use, they can also make creative expression more challenging.
Seance 2.0 is being integrated into various platforms like Polo AI, Runway ML, Cap Cut, and is available through desktop and web applications. For new users, there’s a promotion offering 90% off the first month of the pro plan, effectively providing a free month of access. The model is praised for its cinematic quality and ability to capture a specific mood and line delivery. However, the current 15-second generation limit is a significant constraint, though developers anticipate longer generation times in the future as computing power increases.
GPT Image 2: A Leap in Image Generation
Beyond video, OpenAI is reportedly testing ‘GPT Image 2’, a new image generation model that shows incredible promise. While not yet publicly available, internal testing suggests it could rival or even surpass current leading models like Google’s Nano Banana series. Early examples showcase astonishingly realistic and detailed images, particularly in replicating user interfaces and text within images.
Examples include highly accurate Minecraft screenshots, complete with correct fonts and intricate details, and Windows screenshots that are nearly indistinguishable from real ones. The model’s ability to render legible text within complex scenes, such as game interfaces or technical documents, is particularly impressive. It can generate detailed UI elements, logos, and even recognizable likenesses of public figures in specific contexts, like a Fortnite game screenshot. This level of detail and accuracy in text and UI generation is a significant advancement in AI image creation.
Why This Matters
The emergence of Happy Horse and the potential of GPT Image 2 highlight a rapidly accelerating pace in AI development. Happy Horse’s open-source nature means its advancements will likely be shared and built upon by a wider community, potentially leading to faster improvements in AI video. Its strong performance suggests that high-quality, accessible AI video tools are becoming more common.
Meanwhile, GPT Image 2’s capabilities in generating realistic images with accurate text and interfaces could have profound implications for graphic design, content creation, and even software development. The ability to generate complex, text-heavy visuals with such precision could streamline workflows and open up new creative possibilities. The competition between these advanced models, like Happy Horse and GPT Image 2, is pushing the boundaries of what AI can achieve, offering exciting prospects for future applications across various industries.
Source: HAPPY HORSE 1.0! beats Seedance 2.0 on Leaderboards & likely Open! (YouTube)
