Text-to-Video Guide: How to Write Prompts That Actually Work

Write Prompts That Actually Work

Text-to-video is powerful, but the difference between a mediocre clip and a stunning one is usually in the prompt. Here's a practical recipe for writing prompts in LongCat that produce reliable results.

The Five-Part Prompt Formula

A strong prompt covers five dimensions:

1. Subject — what is in the frame? Be specific: "a red fox" beats "an animal".
2. Action — what is it doing? "sprinting through snow".
3. Setting — where and when? "pine forest at dawn, mist".
4. Style — photographic, cinematic, anime, painterly?
5. Camera & lighting — "slow push-in, soft golden hour light, shallow depth of field".

Example: "A red fox sprinting through snow in a pine forest at dawn, cinematic, slow push-in, soft golden-hour light, shallow depth of field."

Practical Tips

• Lead with the subject. Models weight early tokens more heavily.
• Use motion words deliberately: walking, flying, waves, drift, zoom, pan.
• Keep it under ~60 words. More words dilute attention.
• Describe lighting and camera — they're what make AI clips feel "filmed" rather than "generated".
• Avoid negations. "No blur" is weaker than "sharp focus".

Iterate, Don't Perfect

Generation is cheap — run 2–3 variations, pick the best, and refine only what matters. Small changes to one dimension (style, camera) usually change the whole feel.

Quick-Start Settings

• Duration: 5 seconds for your first tests
• Aspect ratio: 16:9 for YouTube, 9:16 for Shorts/Reels, 1:1 for feeds
• Resolution: start high and scale down only if speed matters

Ready to try it? Open the Studio, pick Text to Video, and paste your first prompt.