Technical notes for logic-native video generation.
Field notes, scoring rubrics, and case-study observations for omni-model, Veo 4, and instruction-based video editing workflows.
Temporal coherence in long-take generation
A comprehensive frame-level analysis of how long-take video generation preserves texture fidelity, object identity, and spatial continuity when both camera and subject remain in continuous motion. This study examines the mechanisms underlying stable symbol rendering across extended temporal windows.
Instruction-native editing vs. pixel diffusion
A systematic comparison between instruction-native scene transformation and traditional pixel-level replacement, focused on whether edits preserve physical and semantic constraints. This benchmark evaluates 14 reference clips across multiple editing scenarios to quantify the advantages of logic-aware editing over surface-level visual manipulation.
Logic Score: a frame-level evaluation rubric
A practical, open-source rubric for scoring video outputs where the central question is whether the model follows causal, symbolic, and spatial logic across time. This methodology provides a structured framework for evaluating reasoning capabilities in video generation systems beyond traditional visual quality metrics.
Failure modes in sub-second causal events
A detailed inspection of fast causal events where the model must preserve temporal order, material response, and secondary motion across only a few frames. This study examines the specific challenges of generating physically plausible short-duration events and catalogs the most common failure patterns observed across multiple model versions.
Understanding Gemini Omni: World Model Architecture and Physical Reasoning
Analysis of Gemini Omni's world model architecture and its approach to understanding physical laws through learning-based prediction rather than symbolic rule systems.
PhyGround Framework: 13 Fundamental Physical Laws for Video Benchmarking
Introduction to PhyGround, a taxonomy of 13 core physical laws used to systematically evaluate video generation models' understanding of physics.