A research team at Shanghai University’s MAGIC Lab has published VERTIGO, a novel framework for cinematic camera trajectory generation, in the proceedings of ECCV 2026. Led by Assistant Professor Li Mengtian and co-authored by students Lu Yuwei and Li Feifei, with collaboration from Wang Xi at École Polytechnique, Paris, the study addresses a persistent bottleneck in generative AI video production: the misalignment between mathematically valid camera paths and professional cinematic composition. Current trajectory generation models typically prioritize spatial accuracy over aesthetic quality, often producing frames with subjects cut off or poorly framed. VERTIGO resolves this by implementing a visual preference optimization loop that mimics professional director feedback. The architecture employs a real-time graphics engine to rapidly convert 3D camera movements into 2D visual previews. A vision-language model evaluates these previews using a novel cyclic semantic scoring method, generating natural language critiques that are semantically compared against original text prompts. This feedback drives the refinement of a camera trajectory generator, which is further optimized through post-training via Direct Preference Optimization. To support the framework, the researchers compiled LenScript, a comprehensive dataset featuring 120,000 trajectories and 21.6 million annotated frames across five key cinematography parameters. Evaluations demonstrate that VERTIGO maintains geometric performance on par with leading methods while effectively preventing subjects from exiting the frame. Independent assessments by 34 film industry professionals confirmed marked enhancements in framing stability, prompt fidelity, and visual appeal. The framework establishes a practical, end-to-end pipeline for AI-assisted pre-visualization, short-form video generation, and animation production. By prioritizing visual usability over raw geometric optimization, the research signals a shift toward AI systems that directly support professional creative workflows. Funded by the National Natural Science Foundation of China, the work positions Shanghai University at the forefront of AI-driven visual media research, illustrating how algorithmic precision can be calibrated to meet established cinematic standards.