VI-Bench (Video Prompt Inversion) - Medium (Style and Camera): leaderboard

Metric: Inversion Score (0-1; mean of the Prompt Score, GPT-4o (gpt-4o-2024-11-20) judging the recovered prompt against the original one on subject, action, scene, style and camera, and the Video Score, a Qwen3-VL-8B memory agent and Qwen3.5-9B judge agent comparing the video regenerated from the recovered prompt by the original generator (Wan2.2 or HunyuanVideo 1.5, fixed seed) with the reference, each a 1-5 rating normalized to 0-1; 300 single-shot videos that add visual style and camera behaviour). Source: arxiv.org. Saturation forecast: Around January 2027. 17 models tracked.

Top models

#ModelScore
1Seed 2.0 Pro0.62
2Qwen 3 VL 8B0.57
3GPT-4o (2024-11-20)0.57
4Qwen 3.5 9B0.55
5Keye-VL-1.5-8B0.53
6Qwen 2.5 VL 7B0.5
7InternVL3-8B0.39

Interactive version: theaggregate.ai/benchmark?slug=vi-bench-video-prompt-inversion-medium-style-and-camera · How It Works · Data refreshed daily, snapshot 2026-09-26.