And while OpenAI continues to impress Sora after months of delays, Tencent has quietly dropped a model that is already showing similar results to existing top-tier video generators.
Tencent unveiled Hunyuan Video, a free and open source AI-powered video generator, strategically timed during OpenAI’s 12-day announcement campaign, which is widely expected to include the debut of Sora, its highly anticipated video tool .
“We are introducing Hunyuan Video, a new open source video platform that offers video creation performance that is comparable to, if not superior to, leading closed source models,” Tencent said in its official announcement.
The Shenzhen, China-based tech giant claims its model “outperforms” the Runway Gen-3, Luma 1.6 and “three high-performance Chinese video generation models” based on the results of professional human evaluation.
The timing couldn’t be more appropriate.
Before its video generator — somewhere between the SDXL and Flux era of open source image generators — Tencent released an image generator with a similar name.
HunyuanDit It provided excellent results and improved bilingual text comprehension, but was not widely adopted. The family is completed with a group of Large linguistic models.
Hunyuan Video uses a decoder-only multimedia large language model as its text encoder instead of the usual CLIP and T5-XXL combination found in other AI-powered video tools and image generators.
Tencent says this helps the model follow instructions better, understand image details more accurately, and learn new tasks quickly without additional training — plus its causal attention setting gets a boost from a special symbolic enhancer that helps it understand prompts more comprehensively from traditional models.
It also rewrites the claims to make them richer and increase the quality of their generations. For example, a prompt that simply says “Man walking his dog” could be enhanced by including details, scene setting, lighting conditions, high-quality artifacts, and ethnicity, among other elements.
Free for fans
Like Meta’s LLaMA 3, Hunyuan is free to use and monetize until it reaches 100 million users — a threshold that most developers won’t need to worry about anytime soon.
hunting? You’ll need a powerful PC with at least 60GB of GPU memory to run the 13B parameter model locally – think Nvidia H800 or H20 cards. That’s more vRAM than most gaming PCs have in total.
For those who don’t have a super-fast computer, cloud services are already coming on board.
FAL.ai, is a generative media platform designed specifically for developers Built-in Hunyuan, charges $0.5 per video. Other cloud providers, including repetition or johannes, It also began providing access to the form. responsible Hunyuan video The server offers 150 points at $10, with each video creation costing a minimum of 15 points.
And of course, users can run the model on a rented GPU using services like Runpod or Vast.ai.
Early tests have shown that Hunyuan matches the quality of commercial heavyweights like the Luma Labs Dream Machine or Kling AI. The videos take about 15 minutes to create, producing realistic sequences with natural-looking human and animal movement.
The test reveals a current weakness: the model’s understanding of English prompts may be clearer than its competitors. However, being open source means that developers can now modify and improve the model.
Tencent says its text encoder achieves up to 68.5% alignment rates — meaning how closely the output matches what users requested — while maintaining 96.4% visual quality scores based on their internal testing.
The full source code and pre-trained weights are available for download at github and Face hugging Platforms.
Edited by Sebastian Sinclair
Smart newsletter in general
A weekly AI journey narrated by Jane, a generative AI model.
Source: https://decrypt.co/295199/tencents-new-ai-video-generator-takes-on-openais-sora-for-free