MPF-Net: Exposing High-Fidelity AI-Generated Video Forgeries via Hierarchical Manifold Deviation and Micro-Temporal Fluctuations

TL;DR

MPF-Net tackles the challenge of detecting high-fidelity AI-generated videos by proposing that synthetic content is governed by a deterministic manifold projection, yielding Manifold Projection Fluctuations (MPF). The method combines a Static Manifold Deviation Branch, powered by Large-Scale Vision Foundation Models, with a Micro-Temporal Fluctuation Branch that exploits micro-residual structure via LoRA-tuned backbones and simple feature fusion. Across VidProm and GenVideo benchmarks, the approach achieves state-of-the-art results, particularly for high-quality videos, and demonstrates how temporal fidelity governs MPF detectability through a quantified quality metric. This dual-path framework offers a practical path for robust forensics in the face of rapidly advancing video generation models.

Abstract

With the rapid advancement of video generation models such as Veo and Wan, the visual quality of synthetic content has reached a level where macro-level semantic errors and temporal inconsistencies are no longer prominent. However, this does not imply that the distinction between real and cutting-edge high-fidelity fake is untraceable. We argue that AI-generated videos are essentially products of a manifold-fitting process rather than a physical recording. Consequently, the pixel composition logic of consecutive adjacent frames residual in AI videos exhibits a structured and homogenous characteristic. We term this phenomenon `Manifold Projection Fluctuations' (MPF). Driven by this insight, we propose a hierarchical dual-path framework that operates as a sequential filtering process. The first, the Static Manifold Deviation Branch, leverages the refined perceptual boundaries of Large-Scale Vision Foundation Models (VFMs) to capture residual spatial anomalies or physical violations that deviate from the natural real-world manifold (off-manifold). For the remaining high-fidelity videos that successfully reside on-manifold and evade spatial detection, we introduce the Micro-Temporal Fluctuation Branch as a secondary, fine-grained filter. By analyzing the structured MPF that persists even in visually perfect sequences, our framework ensures that forgeries are exposed regardless of whether they manifest as global real-world manifold deviations or subtle computational fingerprints.

Figures (9)

• Figure 1: Proposed hierarchical dual-path forensic framework based on manifold and MPF Analysis.
• Figure 2: T-SNE visualization of the hierarchical feature spaces. In Branch I (left), while off-manifold samples with significant distortions are clearly deviated, high-fidelity 'on-manifold' samples heavily overlap with the real-world distribution. In Branch II (right), the same on-manifold samples exhibit distinct, separable clustering when projected into the MPF-based feature space.
• Figure 3: Synthetic Videos that lies on Generator's Coarse Manifold ($M_{Gen}$) but deviate from the structure of Real-World Manifold ($M_{real}$).
• Figure 4: Micro-Temporal Fluctuation (MPF) Analysis, contrasts the homogenous, structured residuals of AI synthesis, constrained by a frozen decoder's fixed basis (left), with the heterogeneous, unstructured noise inherent in physical world recordings (right).
• Figure 5: The hierarchical dual-path framework: Video Preprocessing, Static Manifold Deviation Branch and Micro-Temporal Fluctuation.