<p>OpenAI has amazed the world with its latest development - the Sora model, capable of generating realistic videos. However, not everything is so rosy: generating one minute of video takes about an hour of work on a powerful Nvidia H100 GPU. This makes using Sora prohibitively expensive for regular users and bloggers. It is worth noting that the model is based on transformers and diffusion methods. But what is most impressive is that training Sora requires between 4200 to 10500 Nvidia H100 units over the course of a month.</p>
<p>In practical terms, this means that despite the astonishing quality and capabilities of Sora, the widespread adoption of such models could significantly increase the demand for computational power. And while progress in AI video generation is undeniable, at the moment, the cost and resource intensity make it largely inaccessible for the majority.</p>
<p>Interesting analysis of Sora <a href="https://www.factorialfunds.com/blog/under-the-hood-how-openai-s-sora-model-works">here</a></p>;
<p>#ai #text2video #sora</p>
<p>βββββββββ<br>
ΠΡΡΠ»ΠΈ Π Π²Π°ΡΠ΅Π²Π°<br>
βββββββββ</p>