A Provider-Agnostic Protocol and Observational Pilot for Reproducible Evaluation of Browser-Based AI Video Generators

Browser-based AI video generators are difficult to evaluate reproducibly because interfaces, model routing, and outputs change rapidly. This work presents a provider-agnostic protocol for evaluating accessibility, generation controls, provenance, technical media properties, perceptual quality, prompt alignment, temporal consistency, and operational reliability. An observational pilot analyzes 16 publicly displayed showcase videos (four per service) from Flow AI Video, Seed Imagine, Dola AI, and Luma Dream Machine App. The accompanying reproducibility package includes the experiment manifest, audit script, machine-readable results, and summary. The pilot is descriptive and does not claim a controlled ranking of underlying models. The author is affiliated with the four evaluated web properties; this potential commercial conflict is disclosed and bounded through open methods, machine-readable results, and non-ranking claims.