A study of video generation models Kling 4.0, Hailuo 3.0, and Veo Omni Flash 1.1 identified four key elements for defining style: aspect ratio, camera and lens specifications, lighting, and visual defects. Movement is treated as a separate, fifth element.
Testing conducted between October 20 and 23, 2026, showed that models successfully replicate color and light, but often struggle with texture and specific movement types. Out of 20 tested styles, 17 were generated successfully, while three showed partial failures.
The author recommends using model settings for aspect ratios like 21:9 or 4:3 rather than relying on text prompts. For camera movement, standard industry terminology such as 'push in' or 'orbit' is preferred over subjective adjectives to ensure consistent results.
Models like Kling 4.0 and Hailuo 3.0 have specific limitations regarding continuous shots. For instance, a requested 'whip pan' was rendered as a standard cut, suggesting that users should explicitly request a single continuous shot to avoid unwanted scene transitions.
Effective prompts should place style definitions in the first paragraph and scene details in the second. Developers advise against using vague terms like 'cinematic,' favoring technical descriptions of lenses, lighting, and film stock to guide the AI.
