Stars
[ICML 2026] Twins: Learn to Predict Unified Representations with Focal Loss
ShutterMuse, a open-sourced MLLM that provides both photographpher-side and subject-side guidance.
[ICLR 2026] Official code for [EdiVal-Agent Automated, object-centric evaluation for multi-turn instruction-based image editing]
Self-supervised learning for spatial perception
Masked Depth Modeling for Spatial Perception
[ACM MM 2026] PROVE: A Perceptual RemOVal cohErence Benchmark for Visual Media
ECCV2026 -Perceiving Better Moments: Cover Frame Reselection and Enhancement for Live Photos
PixWorld: Unifying 3D Scene Generation and Reconstruction in Pixel Space
CVPR 2026 (Highlight)-Guiding a Diffusion Transformer with the Internal Dynamics of Itself (IG)
Official implementation of "Perceptual Flow Matching for Few-Step Generative Modeling"
[ICLR 2026] Energy-oriented Diffusion Bridge for Image Restoration with Foundational Diffusion Models
Collect super-resolution related papers, data, repositories
[ICCV 2025][Few-Step Student Surpasses Teacher Diffusion] Learning Few-Step Diffusion Models by Trajectory Distribution Matching
Data-Forcing Distillation (DFD): restoring diversity and fidelity in few-step video generation — text-to-video (Wan2.1) & image-to-video (Cosmos), built on NVIDIA FastGen.
[NeurIPS 2025 D&B🔥] OpenS2V-Nexus: A Detailed Benchmark and Million-Scale Dataset for Subject-to-Video Generation
Development code and experiments for training diffusion models from scratch
Official repository for the paper "PixVerve: Advancing Native UHR Image Generation to 100MP with a Large-Scale High-Quality Dataset"
[ICML 2026] The offical code of Diversity-Preserved Distribution Matching Distillation for Fast Visual Synthesis
JLT: Clean-Latent Prediction in Latent Diffusion Transformers
🚀 [ICLR 2026] SenseFlow: Scaling Distribution Matching for Flow-based Text-to-Image Distillation
PiD: Fast and High-Resolution Latent Decoding with Pixel Diffusion
IQA: Deep Image Structure and Texture Similarity Metric
The official code repo of 1.x-Distill, is a stagewise distillation framework for diversity, high-quality and efficient few-step generation, with support for fractional-step inference and MLP-based …
Continuous-Time Distribution Matching for Few-Step Diffusion Distillation👏
Official implementation of Tuna-2: Pixel Embeddings Beat Vision Encoders for Unified Understanding and Generation