You could leverage Pipeline Builder or AIP Logic for this. You can upload media to a media reference property via an Action, and then pass this to AIP Logic and use the LLM board there. Alternatively, you can use the media set that the media was uploaded to in pipeline builder and use the pipeline to feed the description property on the object.
Are you using a vision model (i.e. GPT 41) and can you validate that you are properly sending a media reference to the useLLM board. You can send a media reference by either defining an input of media reference + passing that into useLLM board or by reading an MediaReference object property + passing that into the useLLM board.
It doesn’t seem like you are actually sending an image to the LLM if that is the proper message you get back from it.