How to Build AI Video Prompts and Cinematic Language: A NadouPro Guide to Shots, Lighting, and Dialogue

When generating videos with AI, some outputs look like casual phone footage, while others look like movie trailers. The difference often does not lie in the model itself, but in the prompt—a prompt that clearly specifies the subject, action, scene, cinematic language, lighting, and atmosphere gives the model a basis for execution.

This guide breaks down a structured way to write cinematic video prompts: from the basic six-element framework, to specific wording for camera movement, shot size, and rhythm, and then to advanced techniques for dialogue, character consistency, and multi-shot sequences. All of these methods can be practiced directly in NadouPro video nodes.

The Six-Element Prompt Framework

There is no mysterious formula for cinematic prompts. In essence, they make the staging clear in writing. A practical framework includes six elements:

Element	Purpose	Example Wording
Subject	Clarifies the main focus of the image	A woman in a striped shirt, a perfume bottle placed on a velvet display stand
Action	Describes visible behavior	Walking toward the camera, swirling juice in a glass, turning around and smiling
Scene	Specifies the location and details	An outdoor terrace, a living room, a stone-paved street in an old city, a studio product scene
Shot	Defines shot size, angle, and movement	Close-up, wide shot, low angle, slow push-in, tracking shot
Lighting	Defines the visible quality of light	Natural light, golden dusk light, cool blue night tones, soft haze
Atmosphere	Sets the emotional direction	Tense, peaceful, nostalgic, cyberpunk

Three writing principles:

The subject description should let the model identify the focus at a glance. Do not write vague phrases such as “magical energy.” Instead, write “rotating blue energy particles with an ethereal glow,” so the image has a clear visual target.

The action should describe things the viewer can see. Smoke drifting upward, flames bending in the wind, a runner leaning forward, the camera tracking close to the subject—visible motion language is far more reliable than abstract technical descriptions.

Lighting and atmosphere determine the emotion of the prompt. Haze, rim light, long shadows, reflections on a wet road—when these visible details are written out, the emotional direction of the image is established.

NadouPro’s cinematic prompt feature can help you expand a simple idea into a complete, well-structured cinematic prompt. It is useful when you are not sure how to begin: draft first, then revise according to your own intent.

Camera Movement Instructions: Direct the Image with Cinematic Language

Camera movement descriptions can turn an ordinary image into a cinematic shot. The principle is to use natural language to clearly state what the viewer will see: a push-in shot slowly approaching the subject, a wide shot used to establish the environment, a low-angle reveal shot, a lateral move slowly scanning across a room, or a tracking shot following the action.

Common camera movements and suitable use cases:

Camera Movement Instruction	Prompt Wording	Suitable Use Case
Close-up	Close-up of the speaker’s face	Dialogue, emotion, product details
Wide shot	A wide shot showing the terrace and two characters	Establishing the environment and spatial scale
Push-in	The camera slowly pushes toward the subject	Emphasis, suspense, reveal
Lateral move or tilt	The camera scans across the room, or tilts up to the sign	Showing space or vertical details
Tracking shot	The camera follows close beside the runner	Motion and action continuity

Taking a typical scene as an example, a complete camera movement prompt can be organized like this:

A smooth, controlled dolly push-in shot gradually approaches an elegant female marble statue on a stone terrace. The camera begins from a medium-wide shot and slowly moves forward, while slightly panning to the right and rising upward, gradually transitioning from a lower angle to a heroic low-angle perspective, sequentially revealing the statue’s drapery, serene facial expression, and graceful posture. The camera movement is smooth and stable, with cinematic control. Soft natural daylight, subtle volumetric light and haze, hyper-realistic texture

The more precise the shot-size terms are, the clearer the visual intent becomes: an extreme close-up focuses on the eyes or the texture of a tiny object; a medium close-up from the chest up works well for dialogue scenes; a full-body shot explains movement and clothing; an extreme wide shot establishes the scale of the scene. For composition, you can also add instructions such as “centered,” “rule of thirds,” or “off-center.”

Rhythm descriptions are equally important: a slow push-in creates calm or suspense, a stable tracking shot makes action easier to read, and a quick cut to a close-up reaction shot creates energy and surprise. NadouPro’s camera parameter settings provide more than 300 focal length, depth of field, and cinematic style presets, allowing you to further lock in the shot texture beyond the prompt.

Dialogue and Sound Effects: Write Sound into the Prompt

Audio-video synchronization models can generate dialogue, ambient sound, and sound effects together with the visuals. On the NadouPro platform, Kling 3.0 Omni 4K, Seedance 2.5, Seedance 2.0, and MiniMax H3 all support this capability.

There are three key points when writing dialogue prompts:

Clearly label the speaker. In multi-person dialogue scenes, write who is speaking before each line. Keep the line, speaker, and tone description together; do not scatter them across different parts of the prompt.

Keep character appearance descriptions stable. Clothing, hairstyle, personal props, and the character’s role in the scene should remain consistent throughout the prompt, so the character’s face is less likely to change.

Make ambient sound specific. Writing “the faint low-frequency hum of an air conditioner in the living room” is much more effective than writing “there is ambient sound.” The more specific the location and acoustic details are, the better the sound will match the image.

When you need high-quality standalone narration, NadouPro also has the Mureka audio model, which provides natural voices and rich styles.

Taking a typical scene as an example, here is a multi-shot prompt with shot labels:

Shot 1: Wide shot, a futuristic female pilot walks through a hangar dotted with neon lights toward her starship; Shot 2: Medium shot, she stops and looks at the starship with a determined expression; Shot 3: Close-up, her hand brushes across the cold metal hull. High consistency, cinematic lighting, realistic texture

Character Consistency: Use Reference Materials to Lock Visual Identity

Keeping a character’s face and outfit the same across different shots is a basic requirement for narrative content. Pure text descriptions make this difficult, so the correct approach is to show the model reference materials.

NadouPro’s multi-reference mode supports uploading multiple reference images. Upload a front view, side view, and key expressions of the character, and the model can follow the character’s facial features, clothing details, and overall proportions as scenes and camera positions change. The multi-panel tool can first generate and finalize a character turnaround and a nine-grid expression sheet, then use them as reference sources for all subsequent shots.

The reference workflow works best when the character description in the prompt remains stable and only the scene, camera position, and action change. The same set of reference materials can support dozens or even hundreds of shots across an entire project.

Multi-Shot Sequences: Generate a Complete Narrative at Once

A single shot cannot make a story. In NadouPro, there are two ways to organize multi-shot sequences:

Generate multiple shots together. Seedance 2.0 and MiniMax H3 support generating multiple coherent shots at once. Organize the prompt in shot order: first define the overall scene setup, then write out each shot’s content, shot size, and camera movement one by one, and give each shot a clear narrative function.

3D director stage. When you need precise control over camera positions and character blocking, use NadouPro’s 3D director stage for visual staging—arrange the duration, angle, and motion path of each shot on a multi-camera timeline, then render the video as a whole, with support for exporting videos up to 90 seconds long. This is currently the multi-shot solution with the strongest control, suitable for commercials, short dramas, and other content with higher requirements for shot continuity.

Text in the Image: Make Every Character Clear

Labels on product packaging, storefront signs, subtitles—when text appears in the image, the prompt should state the exact text content, position, medium, and camera distance. The shorter the text, the easier it is to keep readable; lighting and composition instructions can also help emphasize it.

For rendering text in image scenes, NadouPro’s GPT Image 2 performs especially well. Its hyper-realistic photo quality, combined with precise text and layout capabilities, makes it suitable for product posters and product images that include packaging information.

Workflow Recommendations: Turn Prompts into Reusable Assets

The difference between professional teams and casual experimentation lies in whether there is a systematic method for testing and accumulating results.

Systematic variable testing. Start with a simple base prompt and change only one variable at a time: camera movement speed, shot size, or lighting direction. Record the combinations that work well, and gradually build a prompt library suited to your own content types.

Set parameters according to delivery needs. Model selection, resolution, duration, whether to enable audio-video synchronization, and whether to use reference materials should all be determined according to the publishing scenario for each video. Do not process all content with the same set of parameters.

Brand consistency depends on accumulation. Centrally manage reference materials for characters, products, and brand assets. Turn prompts that have been verified as effective into templates, so anyone on the team can produce content along the same visual direction.

FAQ

Q: How long should a prompt be?

There is no fixed length. The standard is that “each element contains meaningful information.” A prompt that uses three sentences to clearly explain the subject, action, and shot will execute better than a prompt filled with a screenful of adjectives. Multi-shot sequences can be appropriately longer, but the description of each shot should still stay focused.

Q: What should I do if the model does not follow the camera movement I wrote?

Check in three steps: whether the camera movement instruction is buried among many other descriptions—if so, move it to the front of the prompt; whether you are asking for conflicting camera movements at the same time—keep only one main camera movement at a time; whether a visual anchor is missing—add a sentence describing the starting and ending states of the shot. You can also use Agent conversational creation in the NadouPro canvas, send it the generated result, and ask it to help adjust the prompt.

Q: Which languages are supported for audio-video synchronized dialogue?

Multilingual dialogue generation is supported. Lines in different languages can also be mixed within the same scene, while lip movements and facial expressions remain natural. The specific range of supported languages is subject to what is displayed on the platform page.

Q: How can I keep the same character consistent across different videos?

Create a standard reference material package for the character: a turnaround, a nine-grid expression sheet, and clothing detail images. Use the same set of materials as multi-reference input for all videos involving that character. The quality ceiling of the reference materials determines the ceiling of consistency, so this step is worth completing carefully.

Explore more features on NadouPro: NadouPro

Contains AI-generated information that may be incorrect. Please verify independently.