The AI landscape used to feel stagnant and uneventful – but that’s changing rapidly.
Everything shifted when OpenAI unveiled ChatGPT and DALL·E. Suddenly, the world saw just how powerful and transformative generative AI could be.
Tech giants quickly took notice. As the market for AI-driven solutions exploded, companies scrambled to respond – either by integrating OpenAI’s tools or developing their own platforms.
From Google and Meta to emerging players, the race to dominate this space is fierce.
And it’s not just a Western phenomenon. In the East, the demand for generative AI is rising just as fast, signaling a global shift towards AI-powered creativity and innovation.
Tencent from China has what it calls DynamiCrafter. This time, the company goes a step further with 'HunyuanVideo.'
The AI-generated video scene has been hopping this year (or twirling wildly, as the case may be).
After the introduction of OpenAI's Sora and its late release, Google's Veo 2, Kuaishou's Kling AI, Luma AI's Dream Machine, MiniMax's Video-01, Runway's Gen-3-Alpha, PIka's Pika 1.5, and more, HunyuanVideo is entering a saturated market.
The competition is tough, and rivals are more than capable, if not very powerful.
HunyuanVideo is arguably not the best in the competition, but what it lacks is compensated by becoming a neural network that is openly distributed.
Unlike the aforementioned models, Tencent has given HunyuanVideo the ability to run locally.
Among the things the AI can do, include:
- Creating cinematic high-quality video, with artistic experience.
- Native camera cuts, and smooth special effects.
- Continuous actions, will full control.
- Chinese-style creations, and Eastern culture emphasis.
- Concept generalization, with any combination.
During its release, HunyuanVideo, which has 13 billion parameters.
According to Tencent in the AI's project page on GitHub, HunyuanVideo is considered the largest among all open-source video generation models.
However, despite generative AI tools are known for being resource hungry, under the right circumstances, some early testers said that they managed to run this HunyuanVideo on just a consumer 24 GB VRAM GPU.
Being able to run locally also means that users can fine-tune the model to the finest details.
As a result of this, the AI can literally generative videos that are unfiltered.
Provided with enough training data, like nudity or pornography in training data to make it better understand human bodies, it can generate videos of anatomically realistic, nude humans.
However, HunyuanVideo is still like other generative AI models, which have common flaws.
What this means, it learns from existing video clips.
While videos of pretty much anything imaginable is available, not everything that is possible is there.
As a result, HunyuanVideo and all other AI models on the market have some degree of trouble with new scenarios not found in their training data.