LLaVA-Interactive: An All-in-One Demo for Image Chat, Segmentation, Generation and Editing

TL;DR

LLaVA-Interactive is a versatile multimodal human-AI interaction system that supports multi-turn dialogues, visual prompts, and integrates existing AI models for visual chat, segmentation, and editing without additional training.

Contribution

It introduces an all-in-one, cost-efficient multimodal interaction prototype combining multiple pre-trained models for diverse visual and conversational tasks.

Findings

Supports multi-turn multimodal dialogues

Enables visual prompts for aligning human intents

Demonstrates diverse application scenarios

Abstract

LLaVA-Interactive is a research prototype for multimodal human-AI interaction. The system can have multi-turn dialogues with human users by taking multimodal user inputs and generating multimodal responses. Importantly, LLaVA-Interactive goes beyond language prompt, where visual prompt is enabled to align human intents in the interaction. The development of LLaVA-Interactive is extremely cost-efficient as the system combines three multimodal skills of pre-built AI models without additional model training: visual chat of LLaVA, image segmentation from SEEM, as well as image generation and editing from GLIGEN. A diverse set of application scenarios is presented to demonstrate the promises of LLaVA-Interactive and to inspire future research in multimodal interactive systems.