After Two Years, We Finally Have a 'Local Sora'

MustBeSomethingThere · reddit · 2026-08-10

The poster expressed that two years after OpenAI first previewed Sora, open-source video models have finally achieved an impressive 'Local Sora' level.

The post includes a highly cinematic video demo and shares the detailed prompt structure:

Visuals: A static close-up shot framing a glass train window looking out at Tokyo suburbs, with a clear reflection of a young woman looking at her phone inside the cabin.
Sound design: Features a continuous low train rumble, rhythmic track click-clacks, and the muffled swoosh of wind.

This demonstrates that current open-source models can handle complex physical reflections while simultaneously generating matching environmental audio.

Original post →

After Two Years, We Finally Have a 'Local Sora'

More from Multimodal