Contrastive Parallel Denoising for Improving Attribute Alignment of Diffusion models
Abstract
Diffusion models (DMs) have emerged as a powerful class of generative models, demonstrating remarkable performance in high-quality image synthesis, particularly in text-to-image generation. Leveraging iterative denoising steps guided by textual input, these models have enabled significant advancements in controllable image generation across a wide range of applications—from artistic content creation to data augmentation and virtual scene generation. However, despite their success, diffusion models continue to face major challenges in semantic alignment, especially when tasked with rendering complex prompts that involve multiple objects, attributes, or intricate spatial relationships. A recurring issue in this setting is semantic misalignment, where generated images fail to accurately reflect the detailed semantics of the prompt. This often manifests in several ways: missing objects, undesired object fusion (e.g., two distinct objects blending into one), or incorrect object-attribute pairings (e.g., a red shirt rendered on the wrong character). Such errors are particularly common in prompts that contain several object-attribute bindings or hierarchical structures. One key contributing factor is the inter-contamination of textual embeddings during the diffusion process. Since most models encode the full prompt into a global representation, distinctions between different semantic components may become blurred, especially in the later denoising steps. To address these…
Read the paper
Related studies
- Learning from First-Generation Qualitative Approaches in the IS Discipline: An Evolutionary View and Some Implications for Authors and Evaluators (PART 1/2)
- Commentary on the clinical and preclinical dosage limits of interstitially administered magnetic fluids for therapeutic hyperthermia based on current practice and efficacy models
- Prosthesis: New Technological Opportunities and Innovative Biomedical Devices
- The history and future of neural modeling for cochlear implants
- Ryanodine receptors: waking up from refractoriness
- Novel Transgenic Rabbit Model Sheds Light on the Puzzling Role of Matrix Metalloproteinase-12 in Atherosclerosis
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.