Diffusion language models intrinsically fail to capture correlations between decoded tokens, which leads to a harsh trade-off between sampling quality and throughput. To solve this issue, we propose DiLaDiff, a variant of masked diffusion language models with three components: (1) a continuous latent space with semantic capabilities, learned by an auto-encoder fine-tuned from an existing masked diffusion language model; (2) a latent diffusion model learning the prior over the encoder distribution; (3) a consistency model distilling the learned prior into a few-step latent generative model.

A common recipe to improve diffusion models at test-time so that samples score highly against a user-specified reward is to introduce the gradient of the reward into the dynamics of the diffusion itself. This procedure is often ill posed, as user-specified rewards are usually only well defined on the data distribution at the end of generation. While common workarounds to this problem are to use an approximate denoiser to estimate what a sample would have been at the end of generation, we propose a simple solution to this problem by working directly with a flow map.

We present Isekai3D SDG (Synthetic Data Generation), a platform for generating diverse 3D gameplay data for world models, computer vision, neural graphics, and gameplay agents. The platform combines reusable environments and characters with modular Unreal Engine plugins. A two-stage workflow records world simulation and then replays it to capture complementary views and rendering passes from the same underlying gameplay. Configurable cameras, resolutions, and frame rates support paired observations with frame-aligned camera, character, and action metadata.