# How to give video input to Qwen-3VL with online serving

**URL:** <https://discuss.vllm.ai/t/how-to-give-video-input-to-qwen-3vl-with-online-serving/1866>\
**Category:** General\
**Created:** [November 5, 2025, 1:12pm UTC](https://discuss.vllm.ai/t/how-to-give-video-input-to-qwen-3vl-with-online-serving/1866 "2025-11-05T13:12:27Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![IdanBibi](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/idanbibi/32/864_2.png) [@IdanBibi](https://discuss.vllm.ai/u/IdanBibi)\
**Post date:** [November 5, 2025, 1:12pm UTC](https://discuss.vllm.ai/t/how-to-give-video-input-to-qwen-3vl-with-online-serving/1866/1 "2025-11-05T13:12:27Z")

</div>

I have qwen3 vl which was served using vllm. I would like to give it video as input and control how many frames it parses. How can i do it?

---

<div class="post-metadata">

**Author:** ![RunLLM](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/runllm/32/233_2.png) [@RunLLM](https://discuss.vllm.ai/u/RunLLM)\
**Post date:** [November 5, 2025, 1:13pm UTC](https://discuss.vllm.ai/t/how-to-give-video-input-to-qwen-3vl-with-online-serving/1866/2 "2025-11-05T13:13:23Z")

</div>

To control how many frames Qwen3-VL parses from a video in vLLM, you should set the number of frames at server startup using the --media-io-kwargs argument, e.g. --media-io-kwargs ‘{“video”: {“num\_frames”: 40}}’. Alternatively, you can set the fps (frames per second) for sampling via --mm-processor-kwargs ‘{“fps”: 4}’ at startup. These options are not currently supported dynamically per request in online inference; they must be set when launching vllm serve. For more details, see [this discussion](https://github.com/vllm-project/vllm/issues/26903) and [the documentation](https://github.com/vllm-project/vllm/pull/10020).

If you need per-request control, you must pre-sample frames externally before sending the video to vLLM, as runtime adjustment via API is not supported yet. Would you like more detail or code examples?

Sources:

- [https://github.com/vllm-project/vllm/issues/26903](https://github.com/vllm-project/vllm/issues/26903)
- [https://github.com/vllm-project/vllm/pull/10020](https://github.com/vllm-project/vllm/pull/10020)

Hint: Mention @RunLLM in the post for followups.

---

<div class="post-metadata">

**Author:** ![IdanBibi](https://yyz2.discourse-cdn.com/flex036/user_avatar/discuss.vllm.ai/idanbibi/32/864_2.png) [@IdanBibi](https://discuss.vllm.ai/u/IdanBibi)\
**Post date:** [November 5, 2025, 1:20pm UTC](https://discuss.vllm.ai/t/how-to-give-video-input-to-qwen-3vl-with-online-serving/1866/3 "2025-11-05T13:20:07Z")

</div>

I get media-io-kwargs not exist when using vllm serve. I am using vllm==0.11.0