Get Zlio on your home screen
One tap, full screen, like an app.
A research paper argues multimodal LLMs' compositional understanding is underused in composed image retrieval and proposes factorized MLLM embeddings to better map reference image plus text into a retrieval embedding.