Summary
Youri van Hofwegen benchmarks five free AI video generation models — Hyo, Cling 3.0, One 2.7, Veo 3.1, and Cedance 1.5 Pro — against the paid benchmark of Cedans 2.5, using a consistent four-angle character sheet, identical prompts, and two distinct test scenes all run inside the Higgsfield platform to control for variable settings across providers.
The first test focuses on realism and motion consistency using a character walking down a neon-lit rainy street. Cling 3.0 wins by combining realistic textures with natural movement; Hyo finishes last due to texture quality. The second test demands accurate lip sync and deliberate hand movements — lighting a candle while reciting a specific Churchill line — where the rankings flip entirely. Cedance 1.5 Pro takes first place for dialogue accuracy, while Cling 3.0 drops to fourth due to visual glitching. Veo 3.1 performs strongly on dialogue but triggered NSFW content flags four consecutive times on an identical start frame and prompt, requiring a new start image to proceed — a hidden production cost that a side-by-side still image comparison would never reveal.
Van Hofwegen averages both test rankings into a single score and maps each model to its free-tier access route, including resolution caps, session limits, and watermark policies. The conclusion is that no single free model dominates both motion realism and dialogue accuracy simultaneously, making model choice depend on the specific production requirement rather than a single overall winner.
📺 Source: Youri van Hofwegen · Published September 13, 2026
🏷️ Format: Comparison
