Image Generation

ORIGEN: First zero-shot 3D orientation grounding in text-to-image generation.

Optimal Stepsize Distillation accelerates diffusion sampling by distilling knowledge from reference trajectories, achieving 10x speedup with minimal performance loss.

Lumina-Image 2.0: A unified & efficient image generative framework, outperforming previous models with only 2.6B parameters.

LOCATEdit refines cross-attention maps with graph Laplacian regularization, achieving precise & localized text-guided image editing without artifacts.

LeX-Art: High-quality text-to-image generation via scalable data synthesis.

ChatAnyone: Stylized real-time portrait video generation with hierarchical motion diffusion model.

Fixing fine-tuned diffusion models! By using richer, unconditional priors, they generate better images and videos.

BIZGEN: Article-level Visual Text Rendering for Infographics Generation

LongTextAR advances long-text image generation via a novel tokenizer, enabling accurate, controllable, and high-fidelity text rendering in images.

Inference-time scaling for flow models enhances alignment with user preferences via stochastic generation and budget allocation.

Visually perfect generations aren’t always optimal! GenHancer finds that subtly imperfect generations can greatly improve vision-centric tasks.

Bottleneck Sampling: Accelerate diffusion models without retraining by cleverly using low-resolution priors for efficient inference!

LSRNA: Super-resolution in latent space enhances image generation with diffusion models, achieving faster speeds and improved detail.

Aligning image generation subtasks: Equivariant modeling boosts efficiency and generalization by leveraging natural visual signal invariance.

Diffusion-4K: Synthesizing ultra-high-resolution images with a new benchmark dataset and wavelet-based fine-tuning that makes 4K image creation more detailed and accessible!

CFG-Zero*: A better Classifier-Free Guidance to improve the image quality and text alignment in Flow Matching models.

RDTF: Efficient animated sticker generation via dual-mask training, outperforming parameter-efficient tuning under constrained resources.

Adaptive Diffusion Models with Minority-Aware Adaptive DPO

URA: Ultra-resolution adaptation made easy! Uses synthetic data & minor weight tuning for efficient, high-res text-to-image diffusion models.

TokenSet: Tokenizing images as unordered sets for dynamic capacity allocation and robust generation, breaking from fixed-position latent codes.