7 papers
VideoMLA: Low-Rank Latent KV Cache for Minute-Scale Autoregressive Video Diffusion
Hidir Yesiltepe, Jiazhen Hu, Tuna Han Salih Meral +4
Long-rollout causal video diffusion has converged on a fixed-size sliding-window KV cache, with recent progress innovating within this layout by changing which tokens occupy the wi…
Aligning Latent Geometry for Spherical Flow Matching in Image Generation
Tuna Han Salih Meral, Kaan Oktay, Hidir Yesiltepe +2
Latent flow matching for image generation usually transports Gaussian noise to variational autoencoder latents along linear paths. Both endpoints, however, concentrate in thin sphe…
Infinity-RoPE: Action-Controllable Infinite Video Generation Emerges From Autoregressive Self-Rollout
Hidir Yesiltepe, Tuna Han Salih Meral, Adil Kaan Akan +2
Current autoregressive video diffusion models are constrained by three core bottlenecks: (i) the finite temporal horizon imposed by the base model's 3D Rotary Positional Embedding…
Contrastive Test-Time Composition of Multiple LoRA Models for Image Generation
Tuna Han Salih Meral, Enis Simsar, Federico Tombari +1
Low-Rank Adaptation (LoRA) has emerged as a powerful and popular technique for personalization, enabling efficient adaptation of pre-trained image generation models for specific ta…
ConceptAttention: Diffusion Transformers Learn Highly Interpretable Features
Alec Helbling, Tuna Han Salih Meral, Ben Hoover +2
Do the rich representations of multi-modal diffusion transformers (DiTs) exhibit unique properties that enhance their interpretability? We introduce ConceptAttention, a novel metho…
MotionShop: Zero-Shot Motion Transfer in Video Diffusion Models with Mixture of Score Guidance
Hidir Yesiltepe, Tuna Han Salih Meral, Connor Dunlop +1
In this work, we propose the first motion transfer approach in diffusion transformer through Mixture of Score Guidance (MSG), a theoretically-grounded framework for motion transfer…