Tuna Meral bio photo

Tuna Meral

PhD Student @ VT • Generative AI, Diffusion & Autoregressive Models • Video + Multimodal Generation

Publications

Below is a comprehensive list of my research publications, including journal articles, conference papers, and preprints.

Below is a comprehensive list of my research publications, including journal articles, conference papers, and preprints.

CD

ConceptAttention: Diffusion Transformer Learns Highly Interpretable Features

Allen Hallberg, Tuna Han Salih, Ben Howes, Peter Yueyang

arXiv 2025

Without requiring additional training, ConceptAttention repurposes the parameters of diffusion transformers to identify meaningful visual concepts.

CI

Conditional Information Gain Trellis

Tuna Han Salih, et al.

Pattern Recognition Letters 2024

The Information Gain Trellis, originally a tool for conditioning a Learning Classifier System's rule set, is extended to support conditional information gain over specific attributes.

CA

CLoRA: A Contrastive Approach to Compose Multiple LoRA Models

Tuna Han Salih, et al.

arXiv 2024

CLoRA is a training-free method that leverages contrastive learning principles to effectively compose multiple LoRA adapters, preserving each adapter's unique visual style characteristics.

MZ

MotionFlow: Zero-Shot Transfer of Flow Fields for Video Diffusion Models

Tuna Han Salih, et al.

arXiv 2024

MotionFlow is a temporal adapter that transfers motion from real videos to generated videos using flow-based guidance.

MA

MotionFlow-Attention: Driven Motion Transfer in Video Diffusion Models

Tuna Han Salih, et al.

arXiv 2024

MotionFlow is a training-free method that leverages attention for motion transfer between videos, adapting the movement patterns while preserving content integrity.

CC

CONFORM: Contrast is All You Need For High-Fidelity Text-to-Image Diffusion Models

Tuna Han Salih, et al.

CVPR 2024

CONFORM is a training paradigm for text-to-image diffusion models that enhances image-text alignment through contrastive learning principles.