The quality of your AI video depends more on your prompt than on any setting you choose. A vague prompt produces a generic result. A specific, well-structured prompt produces something you can actually use.
This guide covers the practical techniques that work on this site and most other AI video generators.
The 4-part prompt structure
Every effective video prompt has four components:
1. Subject
What is in the scene. Be specific about appearance, position, and number.
- Weak: "a person"
- Better: "a woman in a dark blue linen blazer, mid-thirties, standing at a rooftop bar"
2. Action
What is happening. Describe movement, gesture, or change over time.
- Weak: "walking"
- Better: "walking slowly toward camera, looking over her left shoulder, hair caught by wind"
3. Camera
How the scene is framed and how the camera moves.
- Weak: (nothing)
- Better: "medium close-up, slow dolly push-in, shallow depth of field, rack focus from foreground to subject"
4. Mood
Lighting, color, atmosphere, and film treatment.
- Weak: "nice lighting"
- Better: "golden hour backlight, warm amber highlights, cool blue shadows, 35mm film grain, anamorphic lens flare"
Putting it together
Combine all four parts into a single, flowing description:
A woman in a dark blue linen blazer stands at a rooftop bar at golden hour. She turns slowly toward the camera, hair caught by a light breeze. Medium close-up, slow dolly push-in, shallow depth of field, warm amber highlights with cool blue shadows, 35mm film grain.
This prompt gives the model enough information to make specific choices about every frame.
Common mistakes
Mistake 1: Describing a photo, not a video
Prompts like "a cat sitting on a windowsill, sunlight streaming in" describe a still image. For video, add motion and change:
"A cat sitting on a windowsill turns its head to follow a bird outside. Sunlight shifts through the window, casting moving shadows. Close-up, shallow DOF, warm afternoon light."
Mistake 2: Too many subjects
Each additional subject increases the chance of artifacts. Start with one or two subjects. If you need a crowd scene, describe it as an environment rather than listing individuals.
Mistake 3: Contradictory instructions
"Fast-paced action, slow motion" or "dark moody lighting, bright colorful scene" confuse the model. Pick a consistent mood and stick with it.
Mistake 4: Over-length prompts
Most models have an effective prompt window. After a certain length, additional words get ignored or diluted. Keep prompts under 100 words. Front-load the most important details.
Style-specific templates
Cinematic
[Subject in specific clothing/appearance] [action with specific body movement] in [specific location]. [Camera: shot type, movement, lens]. [Lighting: time of day, color temperature, film treatment]. Shallow depth of field, [specific color grade].
Product ad
[Product name] [positioned how] on [surface/background]. [Camera: slow orbit/push-in/pull-out]. [Lighting: studio/natural, direction]. Clean background, [material detail: glass reflection, fabric texture, liquid motion]. Commercial product photography style.
Anime
[Character description in anime style] [dramatic action]. [Camera: dynamic angle, speed line effect]. [Cel-shaded rendering, expressive eyes, dramatic lighting]. [Specific anime sub-style: shonen, slice-of-life, mecha].
Social media
[Hook moment: surprising reveal, satisfying motion, or eye-catching transition]. [9:16 vertical framing, subject centered]. [Fast pacing, bright colors, high contrast]. [Trending aesthetic: cottagecore, dark academia, Y2K, minimalist].
Testing and iteration
Prompt writing is iterative. Your first attempt rarely nails it. Here is a practical workflow:
- Write your initial prompt using the 4-part structure
- Generate a 5-second clip (lower cost for testing)
- Watch the result and identify what is wrong
- Adjust the specific part that needs improvement
- Regenerate
Keep a text file of prompts that worked. Over time, you build a personal library of reliable templates.
Explore more prompts
Browse the AI video prompt library for tested templates across cinematic, product, anime, and social categories. Each prompt includes recommended settings and a one-click "Use this prompt" button.
FAQ
Does prompt order matter? Somewhat. Most models weight the beginning of the prompt more heavily. Put your most important visual details first.
Should I use technical photography terms? Yes, when you know them. Terms like "shallow depth of field," "35mm," "anamorphic," and "rack focus" are well-understood by video models and produce more specific results.
Is this the official model website? No. This site is an independent browser-based creative tool and is not the official model developer site.
