Get Video Embedding for Video Search

Hi, I am wondering how I can extract the video embedding instead of the video QA result from the model. I want to using video embedding for video search, after getting the video embedding result, how to project text embedding and video embedding into same latent space since visual and text and not using the same encoder?

Start by locating the model's video encoder and the code path that returns video QA results; the issue does not name files, tests, or entry points. Determine whether the repository exposes video embeddings and how text and video representations are produced. Done would require a documented, tested way to obtain compatible embeddings for video search.

Written by the indexing model from the issue text.