NExT-GPT

NExT-GPT uses a large language model as the hub, connecting encoders and generators for different modalities. A user can input text, images, videos, or audio, and the system can output another modality or multiple modalities.

Its point is pushing multimodality beyond image-text question answering toward fuller any-to-any conversion.

Multimodality is one of the core directions for the next stage of large models. NExT-GPT explores the orchestration problem early: how specialized models can coordinate around an LLM instead of retraining one giant model for every input-output pairing.

That path matters for research and gives application builders a composable architecture reference.

NExT-GPT shows that NUS has globally visible work in multimodal foundation-model research. It is not a local Singapore application project, but a sample of Singapore academia participating in global model-paradigm competition.

This page is a place to keep adding citations, follow-on models, industrial translation, and links to other NUS multimodal teams.