# Out of memory / stuck generations (/docs/troubleshooting/out-of-memory)



## The first generation of a session is very slow [#the-first-generation-of-a-session-is-very-slow]

The model is being read off disk before any generating starts, which is a different problem from running out of memory.

**Fix:** [The first generation is very slow](/docs/troubleshooting/first-run-slow).

## A video generation fails immediately at 0% [#a-video-generation-fails-immediately-at-0]

This usually means the GPU ran out of memory before it could start.

**Fix:** Close other GPU-heavy apps, choose a lower resolution or shorter clip, or pick a lighter preset. After an out-of-memory error, restart the engine before trying again: click the **Local Engine** item on the rail's bottom system strip and choose **"Restart"**.

## A generation sits "stuck" near 80% for a long time [#a-generation-sits-stuck-near-80-for-a-long-time]

On cards with limited VRAM, the final decode step can fall back to the CPU, which is slow but usually finishes. It can take a long time on long or high-resolution clips.

**Fix:** Let it finish if you can, or next time use a shorter duration / lower resolution. Cloud generation avoids this entirely.

## A video generation crashes part-way through, or the whole PC freezes [#a-video-generation-crashes-part-way-through-or-the-whole-pc-freezes]

That is a different problem from running out of VRAM: Windows has run out of room to promise memory, not the GPU. The tell is that lowering the VRAM profile or turning on Low Memory Mode does not help, or makes it worse.

**Fix:** [Increase your Windows page file](/docs/troubleshooting/page-file).

## A cloud generation timed out [#a-cloud-generation-timed-out]

Very heavy settings (long duration at the highest resolution) can exceed the time limit.

**Fix:** Reduce the duration or resolution. If you were charged for a generation that produced nothing, it is automatically refunded; if not, contact support.