Generate Ultra-Realistic Cinematic Stunt Videos

Creating a cinematic stunt video requires precise prompt engineering to maintain identity consistency and realistic physics. This example demonstrates how to generate an 11-second ultra-realistic sequence using a reference image.

The core instruction uses the uploaded reference image as the exact identity and appearance reference for the seated person. The scene is set in a dry, open field with dusty uneven ground, featuring a person sitting calmly on a black folding chair.

Key visual elements include an all-black outfit, black sunglasses, and crossed legs. The prompt mandates preserving exact face, hairstyle, skin tone, and body proportions throughout the entire video to prevent identity drift or deformation.

The narrative unfolds in distinct time segments to ensure physical realism and camera consistency. Each phase builds upon the previous one while maintaining the subject's calm demeanor amidst dynamic action.

• 0-3 seconds: The person remains seated, calm, and still, establishing the baseline pose and environment.
• 3-6 seconds: A RED luxury SUV approaches from behind at high speed, kicking up natural dust.
• 6-8 seconds: The SUV hits uneven ground and jumps toward the camera with believable suspension and tire movement.
• 8-9 seconds: The vehicle lands naturally, showing visible suspension compression and realistic dust impact reactions.
• 9-11 seconds: The SUV passes the seated person while dust spreads, keeping the subject unchanged.

Technical specifications emphasize photorealistic professional-camera footage with natural sunlight and realistic shadows. Cinematic depth of field, subtle motion blur, and detailed 4K quality are required to enhance the visual fidelity.

Negative constraints are critical for controlling the output. The prompt explicitly forbids extra people, changing clothes, floating vehicles, unrealistic physics, text, watermarks, or any CGI/cartoon appearance.

Maintaining consistent camera perspective, composition, and environment ensures the final video looks like continuous professional-grade footage rather than a series of disjointed clips.