The quality of AI video generation is determined by the prompt. Even with the same engine and the same credits, how you write the prompt drastically changes the result. Here are effective prompt-writing principles distilled from thousands of generation experiments.
Basic Structure: Subject → Action → Background → Mood → Camera
A good video prompt includes five elements in order: subject (who/what), action (what they're doing), background (where), mood (what feeling), and camera style (how it's shot). Example: 'A chef (subject) slicing fresh vegetables (action) in a modern kitchen with warm lighting (background), cozy and professional atmosphere (mood), close-up cinematic shot with shallow depth of field (camera)'.
Specificity Is Everything
'Beautiful woman' produces far less predictable results than 'a 30-year-old Korean woman with short black hair wearing a beige trench coat'. AI models understand concrete descriptions much better than vague adjectives. Specify colors, materials, shapes, and quantities. That said, cramming in too many details at once can cause the model to ignore some — focus on 5–7 core elements.
Use Camera Terminology Actively
AI video models understand real cinematography terms. Use these based on context. Shot size: extreme close-up, close-up, medium shot, wide shot, aerial shot. Camera movement: slow pan left, tilt up, zoom out, tracking shot, handheld. Lens effects: shallow depth of field, bokeh background, wide-angle distortion, anamorphic lens flare. Combining these lets you direct the AI like a film director giving shot instructions.
Remove Unwanted Elements with Negative Prompts
Some Skaper nodes support Negative Prompts. Enter elements you don't want in the video here. Example: 'blurry, distorted faces, unrealistic proportions, watermark, text overlay, overexposed, low quality'. For portrait videos especially, adding 'deformed hands, extra fingers, unnatural eyes' to the negative prompt can significantly reduce the unnatural body rendering typical of AI.
Reference a Style
You can reference films, directors, or cinematographers that capture the style you want. Example: 'cinematic style of Wong Kar-wai', 'shot like a Wes Anderson film', 'color grading similar to Blade Runner 2049'. These references won't always reproduce perfectly, but they're effective for establishing overall mood, color, and composition. For advertising, try 'inspired by Apple product ads' or '90s Levi's commercial aesthetic'.
Iterate and Vary
Prompt engineering is fundamentally iterative. Rather than expecting a perfect result immediately, start with a basic prompt and change one element at a time, observing how results shift. When you hit a result you like, save that prompt using the node memo feature — reusing it in similar projects saves both generation time and credits.