Video Understanding Support

[loveunk]

Qwen2.5-VL supports video understanding, when will LM Studio supports uploading video and do inference?

[Sana-A-E]

I've been waiting for this too. It's been a while and still no video?

[rudolphos]

Same question, LM Studio 0.3.31 (Build 5) currently shows "This file type is not officially supported" for mp4,webm files

[DCVirtualCosmos]

More now than ever with Qwen3 VL, the best visual language model open source.

[iEddie-cmd]

when video?

[HavenCTO]

still no video understanding

[Sana-A-E]

It's been a year now and still no video? Are you guys ever going to add it?

[nobody-forever]

still no video

[delphijb]

Yes, I'm very interested. It would put LM Studio ahead of the curve (although Ollama already supports it).

It's frustrating to have models like Qwen3.5 and not be able to do video analysis with them, even though they are natively compatible! Not to mention other even more multimodal models (Qwen Omni, etc.).

[delphijb]

@ryan-the-crayon @rugvedS07 ?

[wzgrx]

wait

[nobody-forever]

wait

people waited a year since the start of this issue, still no response

[nobody-forever]

@ryan-the-crayon @rugvedS07 answer maybe?

[marwanlhabti5-coder]

Unfortunately we need alternative

[nobody-forever]

Unfortunately we need alternative

which alternatives you can suggest?

[marwanlhabti5-coder]

Unfortunately we need alternative

which alternatives you can suggest?

the best alternatives i use right now is ollama server with local api calling i use qwen3-vl (ollama do not support video yet but i use tricks) i convert the video i want into frames and send it about 2 minutes of video frames at once and continue like this 2 minutes by 2 minutes until it process the full video frames and its surprisingly well, its able to understand the timing and treat the images frames batch as 1 video and able to tell what changes, dont forget to increase the context limits in ollama settings based on your gpu so the it dosent give you error out of context, and try to prompt it to describe what changes during the images and not to treat each frame at like image at its own and like this you get better understanding model of time, if you have a video like 60 fps try to convert it to 1 fps or 2 fps as other frames are unecessery compute