Paper Overview
Field: Computer Vision (CV) Authors: Zixuan Li, Haokun Lin, Yicheng Xiao arXiv: 2506.14664
Summary
Unified multi-modal large language models (MLLMs) have achieved strong text-to-image generation quality, but still struggle with structure-aware prompt following, where object counts, spatial relations, attribute bindings, and coarse layouts must be preserved. The authors attribute this limitation in part to the entanglement of structural planning and appearance rendering within a single conditioning stream.
To address this issue, they propose Implicit Visual Chain-of-Thought (IV-CoT), a latent visual reasoning framework for query-conditioned image generation. IV-CoT decomposes the visual conditioning queries into a structural-to-semantic cascade:
- Structural queries first form a latent visual plan.
- Semantic queries then render appearance conditioned on this plan.
- Identifies the entanglement of structural planning and appearance rendering as a key limitation of unified MLLMs in text-to-image generation.
- Introduces a structural-to-semantic cascade of latent visual queries enabling implicit CoT reasoning.
- Uses training-only sketch supervision; no sketch extraction or intermediate decoding needed at inference.
- Single forward-pass inference with strong results on GenEval and T2I-CompBench.
To guide the structural queries, the method introduces training-only sketch supervision, encouraging them to capture structure from sketches without requiring sketch extraction or intermediate decoding at inference time. IV-CoT performs implicit chain-of-thought reasoning in a single forward pass and achieves excellent results on GenEval and T2I-CompBench.
Visualizations and analysis demonstrate that the learned structural and semantic queries play complementary roles in structure-aware generation.
Key Points
*Auto-collected on 2026-06-25.*