A useful LTX prompt reads like a shot description, not a bag of style tags. Establish the shot and setting, describe an action in sequence, then add the character details, lighting, camera motion, ambient sound, music, and dialogue that the shot actually needs.
Use present-tense verbs and concrete physical cues. A close-up needs facial detail, hands, texture, and vocal delivery; a wide shot needs spatial relationships, light sources, and the placement of people or objects. Rather than naming an emotion alone, show it through a trembling lip, an avoided glance, or a pause before speech.
Camera directions need an endpoint. Do not stop at “the camera pans right”; say what the pan reveals and how that new image connects to the subject. Keep dialogue inside quotation marks and specify the requested language, accent, or delivery when relevant. Give ambient sound a source and let music enter at a defined moment.
Before generating, read the prompt as a timeline: clear opening frame, continuous action, camera movement with a destination, and audio with a source. Refine it from the output instead of merely swapping adjectives. The full supplied source document and examples are appended to the Simplified Chinese version.
