Partner
FilmArena
AI model ranking · technology partner · Technology
FilmArena ranks generative image and video models by blind pairwise voting instead of marketing claims. Viewers compare two outputs from the same prompt without knowing which model made them, and a Bradley-Terry model turns those votes into Elo ratings, over a million votes so far. Scores break down by emotion, multi-character scenes, heavy movement, object handling and world building. Built by Koyal. The leaderboard has separate tabs for video, video edit and image edit, lets the viewer filter realistic or animated styles, and sorts by Elo or by win rate. For each model it shows the number of wins, matches and votes, and the site calls the ranking community-driven and based on open data. Besides the score for overall quality, it measures visual fidelity and prompt adherence as quality, temporal coherence for video or style stability for images as consistency, and generation throughput as speed. A published methodology report explains how image and video models are compared through blind side-by-side preference, and a sandbox lets anyone upload their own images, compare models side by side and generate videos or edits with the top-ranked ones. FilmArena says its team includes researchers from Meta, MIT and CMU.
