> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getimpala.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Video

> Send video to vision-language models in batch chat-completion requests.

Vision-language models on Impala accept video as input. Send it as a reference to an object in S3, or inline as base64.

The inference engine splits the video into frames and encodes them with the prompt. You can set the frame rate per request with `mm_processor_kwargs`:

```json theme={null}
{"mm_processor_kwargs": {"fps": 2}}
```

For S3 access requirements, see [Images](/pass_images_with_image_url#s3-access). The same bucket setup applies.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.