Feature Request]: Integration of TIPSv2 Vision Encoder

Currently available open-source workflows heavily rely on standard CLIP or SigLIP text/image encoders for semantic alignment. While excellent for global prompt-vibe tracking, these encoders suffer from spatial blindness

I would love to request core support or an ecosystem framework for Google DeepMind's newly released TIPSv2 foundation vision encoder (google/tipsv2-b14), built on the iBOT++ architecture.

Start by reviewing the existing CLIP and SigLIP encoder integrations and the google/tipsv2-b14 model requirements. Determine whether the goal is core support or an ecosystem framework, then define an integration path and validation criteria for spatial-semantic alignment.

Written by the indexing model from the issue text.