Driving Unreal Engine from language, so a world simulation exposes the objects and transitions you want to edit
AspireAI community · arXiv
Controllable world simulation needs more than visually plausible frames: scenes must retain persistent state, coordinated actions and editable temporal structure. Generative world models synthesise flexible visual output but do not always expose the objects and transitions needed for precise revision, while graphics engines give executable state and direct control but require specialised knowledge of assets, geometry and interdependent APIs to translate language into valid operations. CineU, short for Cinematize from Unreal, is an agentic system for language-driven world orchestration and cinematic choreography inside Unreal Engine, treating an Unreal project as a persistent, inspectable world representation while agent-generated code specifies operations whose effects can be executed and verified. It combines hierarchical orchestration, cascaded asset retrieval and geometry-constrained placement, capability-aligned code synthesis, and multi-stage verification with localised repair, with shared scene and timeline context coordinating entities, actions, camera trajectories and shot boundaries and supporting dependency-aware revisions. Experiments on scene construction, selective editing, multi-shot cinematography and component ablations show how specialised execution knowledge and engine feedback improve task completion and targeted recovery. The limitation is that the system is tied to one engine's asset and API surface, so the interesting part, deciding which parts of a scene to expose and how to sequence them, has to be re-implemented for any other target.
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.