Generate Videos on Your Computer with Pyramid Flow

AI-generated videos with 768p resolution and 24 FPS

Several AI models can already generate videos, but most have major issues like low resolution, high running costs, or being limited to API access. Open models that can create high-resolution videos on consumer hardware are still uncommon.

Recently, Pyramid Flow was released. This model can do fast video generation at 768p resolution and 24 FPS. It accepts both text prompts and text+image prompts to guide the generation.

In this article, we will first review the technical report on Pyramid Flow to understand how it works and the technology behind it. Next, we will generate short videos using a consumer GPU. I used an A40 from RunPod (referral link) but an RTX 3090 would be enough for lower-resolution videos. We'll walk through the code step-by-step, using both image and text prompts.

I’ve also created a notebook that provides all the steps for generating videos. You can find it here: