Search

TwinFlow is a new way to make big image models draw great pictures in just one step instead of 40–100 steps.

PromptRL teaches a language model to rewrite prompts while a flow-based image model learns to draw, and both are trained together using the same rewards.

RecTok is a new visual tokenizer that teaches the whole training path of a diffusion model (the forward flow) to be smart about image meaning, not just the starting latent features.

Flow Matching is like teaching arrows to push points from a simple cloud (source) to real pictures (target); most people start from a Gaussian cloud because it points equally in all directions.

This paper introduces Self-E, a text-to-image model that learns from scratch and can generate good pictures in any number of steps, from just a few to many.

This paper shows that when teaching image generators with reinforcement learning, only a few early, very noisy steps actually help the model learn what people like.

SAM Audio is a new AI that can pull out exactly the sound you want from a noisy mix using text, clicks on a video, and time ranges—together or separately.

DenseGRPO teaches image models using lots of small, timely rewards instead of one final score at the end.

DINO-SAE is a new autoencoder that keeps both the meaning of an image (semantics) and tiny textures (fine details) at the same time.

Short videos are easy for AI to make sharp and lively, but long videos need stories and memory, and there isn’t much training data for that.

Text-to-image models using GRPO used to give the same final reward to every step, which is like giving the whole team the same grade no matter who did what.

SARAH is a real-time system that makes virtual characters move their whole bodies naturally during a conversation while knowing where the user is.