Vertigo Vertigo: Reconstructing a Cinematic Ideal through its Predictive AI Double

2026-06-29 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision & Pattern Recognition, Multimedia · Depth: Expert, short

Summary

Vertigo Vertigo is an AI-driven reconstruction of Alfred Hitchcock's 1958 film "Vertigo," generated using only 2.78% of the original frames. Researchers Adam Cole and Mick Grierson employed a large video diffusion model for first-last frame interpolation to predict the intervening sequences. Computational analysis and critical feedback confirmed remarkable structural fidelity, with 73.1% of frames recognizable as plausible renditions and only 3.6% failing catastrophically. This high fidelity suggests that classical cinematic norms are deeply compressed within the model's latent priors. Aesthetically, the reconstruction presents an unstable overlay, creating a "21st-century vertigo" that questions authenticity. The work argues that generative media accelerates cinema's inherent logic of desire and false authenticity, rather than representing a paradigm shift.

Key takeaway

For creative technologists and media theorists exploring generative AI, this project demonstrates how AI models encode and extend classical cinematic conventions. You should consider using sparse keyframe interpolation with diffusion models to reconstruct or reinterpret existing media, revealing latent aesthetic priors. This approach offers a powerful tool for critical analysis of AI's relationship to art history and perception, prompting new questions about authenticity.

Key insights

Generative AI, specifically video diffusion models, encodes and extends classical cinematic norms and themes of artificiality.

Principles

• Cinematic norms are deeply compressed in AI latent priors.
• Generative media accelerates cinema's logic of desire.
• Sparse keyframes enable high-fidelity video reconstruction.

Method

First-last frame interpolation using a large video diffusion model, anchored by 2.78% of original film frames, to predict intervening sequences.

In practice

• Reconstruct classic films with minimal source data.
• Analyze AI's encoding of artistic conventions.
• Create new aesthetic experiences from existing media.

Topics

• Video Diffusion Models
• AI Film Reconstruction
• Cinematic Norms
• Generative Media Aesthetics
• First-Last Frame Interpolation
• Hitchcock's Vertigo

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, Creative Technologist

Related on AIssential

• Novel View Synthesis as Video Completion — Video diffusion models can be adapted for sparse novel view synthesis by treating it as low frame-rate video completion.
• CineCap: Structured Reasoning with Spatio-Temporal Anchors for Cinematographic Video Captioning — CineCap uses structured reasoning and reinforcement learning to generate accurate cinematographic video captions.
• Toward More Controllable AI Video Editing: An Early Research Exploration at Netflix — Netflix's Vera and VOID models advance controllable AI video editing by isolating changes and ensuring physical plausibility.
• Rein3D: Reinforced 3D Indoor Scene Generation with Panoramic Video Diffusion Models — Rein3D integrates 3D Gaussian Splatting with panoramic video diffusion to restore and refine 3D indoor scenes from sparse inputs.
• Gen4U: Unifying Video Generation and Understanding via Diffusion — State-of-the-art video diffusion models capture both low-level geometry and high-level semantics, enabling unified generation and understanding.
• Video Reconstruction using Diffusion-based Image-to-Video Generation with Trajectory Guidance — Trajectory-guided diffusion models can reconstruct missing video frames by integrating external sensor data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.