Tag
This technical report presents Qwen-Image-2.0-RL, a post-training pipeline using reinforcement learning from human feedback and on-policy distillation to enhance visual quality and instruction-following in image generation and editing tasks.
This paper introduces Prob-BBDM, a probabilistic Brownian Bridge Diffusion Model for efficient and high-quality MRI sequence synthesis from 2D axial slices, achieving up to 88.46% SSIM and 26.09 dB PSNR with only 4 diffusion steps, and demonstrating clinical utility in tumor segmentation.
NVIDIA announces DFlash, an open source block diffusion model for speculative decoding that achieves up to 15x higher inference throughput on Blackwell GPUs while maintaining interactivity.
This paper presents TryOnCrafter, a novel framework for camera-controllable video virtual try-on that uses a renderable 4D try-on proxy and DiT-based video generation to achieve omnidirectional viewpoint exploration, overcoming the limitations of existing methods that depend on fixed source camera trajectories.
Nvidia claims a 15x speedup in text generation using a diffusion model, generating entire blocks at once.
A diffusion model that can transform any image into an interactive, playable hallucination, running locally on user hardware.
Krea 2 is a 12-billion parameter text-to-image diffusion model released open-weight on Hugging Face, with Raw (base) and Turbo (post-trained) checkpoints available.
Vera is a layered diffusion model for video editing that preserves source content by generating edit layers and alpha mattes, using a Mixture-of-Transformers architecture.
Inception Labs released Mercury 2, a diffusion language model that generates roughly 1,000 tokens per second and outperforms Google's DiffusionGemma on the AIME 2026 benchmark with a score of 90% versus 69.1%, though DiffusionGemma is free and open-weight while Mercury 2 is a paid, closed-weight API model.
Krea released Krea 2 Turbo, a 12-billion parameter text-to-image diffusion model, available open-weight on Hugging Face with support for multiple inference libraries.
DiRecT introduces a training-free algorithm for safe diffusion-based planning that enforces constraints only on final clean trajectories using receding-horizon denoising, improving safety and performance over existing methods.
Introduces Decoupled Latent Optimization (DLO) for full waveform inversion, which relaxes latent optimization into a quadratic-penalty objective, outperforming classical and diffusion-based methods on benchmarks while preserving smoothed-velocity initialization.
Track2View generates novel camera viewpoints from videos by conditioning a video diffusion transformer on paired 3D point tracks, achieving state-of-the-art visual quality and significant reductions in rotation and translation errors.
APCyc is a target-aware generative framework that designs cyclic peptides with controlled physicochemical properties by explicitly modeling cyclization patterns and using Bayesian posterior guidance.
Pythagoras-Prover is a compute-efficient family of Lean theorem provers that achieves strong performance using curriculum supervised fine-tuning and a novel Augmented Lean Formalisation technique. The 4B model surpasses DeepSeek-Prover-V2-671B at pass@32 on MiniF2F-Test, and the 32B model sets a new state-of-the-art among open-source provers.
DeepMind researcher Brendan O'Donoghue provides an in-depth introduction to text diffusion models, which generate text through iterative denoising. Compared to autoregressive models, they offer lower latency but limited throughput, and demonstrate unique advantages such as self-correction and dynamic computation.
Google released DiffusionGemma, an experimental open-source diffusion model for text generation that achieves 4x speed boost over autoregressive models, optimized for local processing.
This paper introduces Lip Forcing, the first autoregressive diffusion method for real-time video-to-video lip synchronization. By distilling a 14B teacher into causal students and using only two denoising steps, it achieves 31 FPS streaming at 1.3B scale, 17.6x faster than same-scale bidirectional models.
ControlNet author Min Shen has open-sourced the FramePack video generation model, which requires only 6GB of VRAM to run a 13B model, generates a 1-minute 30fps video, takes 1.5 seconds per frame on an RTX 4090, and comes with a one-click Windows package.
The paper introduces MilliVid, a method for improving long-range consistency in video generation by using a multi-scale autoencoder to compress frames into hierarchical tokens and then generating them with a coarse-to-fine diffusion model, outperforming baselines on Minecraft videos.