How to Write Prompts That Get Better AI Results

Be Specific About the Subject

Vague: "a person walking in a city." Better: "a woman in her 30s in a navy coat walking through a busy Tokyo street at dusk, viewed from behind." The more specific the subject, the less the model guesses.

Describe Style Explicitly

Don't assume the model knows "cinematic" means shallow depth of field with warm color grading. Say exactly that: "shallow depth of field, warm color grade, 24mm lens, golden hour lighting."

Specify Camera and Framing

• Focal length: 24mm (wide), 85mm (portrait), 200mm (compressed)
• Angle: eye-level, low angle, bird's eye, Dutch tilt
• Shot type: close-up, medium shot, wide establishing shot
• Movement: static, slow dolly in, pan left

Keep Continuity Across Scenes

For multi-scene videos, keep character description, lighting style, and color palette consistent in every prompt. Copy the visual anchor phrases from scene to scene.

Avoid Overloading

More than 4–5 distinct requirements in one prompt usually causes the model to drop or blend some of them. Prioritise the most important visual elements.

Iterate

Treat the first generation as a draft. Adjust one variable at a time to see what changes — don't rewrite the whole prompt when you only need to tweak the lighting.