InsertFuse framework enhances multi-category image insertion with novel distillation and conditioning

Researchers have introduced InsertFuse, a novel framework designed for reference-guided image insertion across multiple categories. The system decouples category-specific learning from cross-category consolidation by training specialized "experts" and then using Insertion On-Policy Distillation (IOPD) to merge their capabilities into a single model. To enhance spatial control and reference fidelity, InsertFuse incorporates Token-Aligned Geometry Conditioning (TAGC) and Region-Balanced Flow Matching, along with Reference CFG to strengthen visual guidance. Experiments on the AnyInsertion benchmark and a new multi-category dataset show that InsertFuse achieves state-of-the-art performance in generation quality and reference adherence. AI

IMPACT This framework could improve the quality and control of AI-generated images, particularly in applications requiring precise reference-guided insertions.

RANK_REASON This is a research paper detailing a new framework and methods for image insertion. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• AnyInsertion
• arXiv
• InsertFuse
• Insertion On-Policy Distillation
• IOPD
• Reference CFG
• Region-Balanced Flow Matching
• Token-Aligned Geometry Conditioning

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →