Squeezing Capacity from Multimodal Large Language Models for Subject-driven Generation
Conditioning diffusion models on multimodal large language models with VAE identity conditioning and dual-layer aggregation improves subject-driven generation by balancing semantics with identity preservation.
Published 2026Sydney Poster Session 2 · Tue, Dec 8, 5:00 PM–8:00 PM local time · Hall 1-4▲ 7 on Hugging FacearXiv ↗OpenReview ↗
Only vote on papers you've read. Sign in with GitHub to vote.
It brilliantly marries MLLM cross-modal reasoning with VAE identity injection to cure copy-paste artifacts, though its dual-layer aggregation and multi-stage balance still await rigorous ablation and automated metric proof.
Abstract
Subject-driven image generation aims to synthesize new images that preserve the identity of the given subject while following textual instructions. Existing approaches often encode text and reference images separately. This limits cross-modal reasoning abilities and causes copy-paste artifacts. Recent frameworks that connect multimodal models and diffusion models improve instruction following, but largely overlook identity preservation. To address these limitations, we condition diffusion models on Multimodal Large Language Models (MLLMs) that jointly encode text and reference images, and augment it with VAE-based identity conditioning. A novel Dual Layer Aggregation (DLA) module is designed to aggregate multi-level MLLM features for optimal conditioning, and a multi-stage denoising strategy is applied to progressively balance the semantic information from MLLM and fine-detail identity from VAE during inference. Extensive experiments demonstrate that our approach harmonizes multimodal understanding with identity preservation, mitigates copy-paste issues, and achieves superior performance regarding human preference on subject-driven image generation. Our project website is available at https://zsh2000.github.io/squeeze-mllm-subject-gen/.