How to Manage Characters and Props with Multi-reference? How to Keep Multi-Element Consistency in AI Video? NadouPro Multi-reference Video Tutorial

Key Takeaways

• A reference-driven workflow connects prompts with assets: character references manage people, image references guide style and scenes, and video references guide motion and voice characteristics

• Each reference asset needs a clear role: which one defines the character, which one defines the product, and which one guides the scene; state each role clearly in the prompt

• Prompts should still read naturally: name the character, product, prop, scene, action, and speaker so each reference asset can do its job

• Audio-video sync models support multilingual dialogue; label the speaker, language, lines, and tone item by item so multi-character scenes do not get confused

• For multi-shot sequences, plan the shots before writing the prompt: define duration, shot size, angle, narrative content, and camera movement one by one

Four Types of Reference Inputs, Each with Its Own Role

Models integrated into NadouPro, such as Kling 3.0 Omni 4K, are naturally suited to reference-driven creation. There are four common types of reference inputs:

• Element reference: a character, product, prop, or any subject that needs to appear repeatedly; describe how it appears in the frame and what role it plays

• Image reference: guides subject features, style, or scene direction; specify in the prompt which visual dimension the image should guide

• Video reference: used when you need to preserve both the character’s visual traits and voice characteristics

• Sound binding: binds the voice direction to a character element, with speaker, lines, language, and tone labeled together

Reference-driven prompts should still read naturally. Name the character, product, prop, scene, action, and speaker so the reference assets have clear creative roles.

For recurring characters and hosts in serialized content, use reference assets to lock the look, then directly describe the scene action, dialogue, and shot relationships.

Practical Ways to Use Reference Assets

Marketing scenarios: product reference images can lock the shape, color, and visual recognition of the featured product, then place it repeatedly in different scenes so the product image stays consistent.

Cinematic sequences: clear action, speaker labels, and visual references are the foundation. In dialogue scenes, name each character, each line, the delivery, and the gesture or expression paired with that line.

Multi-shot planning: write scene coverage and shot information in the prompt so the model can adjust camera position and composition accordingly; when full control is needed, define duration, shot size, angle, narrative content, and camera movement shot by shot. NadouPro’s Storyboard can split a script into a shot list with one click, and each shot can be executed separately.

How to Describe Motion So It Feels Real

Realistic movement depends on clear action language: speed, direction, body posture, contact with the environment, and how the camera follows the subject.

• Falling or jumping: describe takeoff, the body arc, landing, and contact so the action has readable stages

• Wind or fabric: describe what is moving and in which direction, so clothing, hair, leaves, and smoke have a visible response

• Water or dust: describe ripples, splashes, trails, and rising dust so the environment reacts to the subject

• Camera relationship: describe Dolly In, tracking, panning, tilting, or composition so the audience clearly understands the meaning of the movement

For objects and natural elements, directly describe what the audience should see: falling leaves, fabric pulled by the wind, reflections on wet ground, ripples after footsteps, or a hand holding a prop.

Environmental Details and Lighting

Environmental details make a scene believable: visible shadows, reflections, mist, rain, wind, surface contact, and how these details interact with the subject.

Write natural elements as visible effects: how the rain falls, which way the wind blows, how the grass moves, where it appears, and how it affects the subject and environment. Write liquids, dust, smoke, and reflections as visual behavior—ripples spreading across the water, dust rising from the ground, smoke drifting past a porch, and reflections stretching across a wet road.

Audio-Video Sync and Multi-Character Dialogue

Kling 3.0 Omni 4K supports audio-video sync generation, producing dialogue, sound effects, ambient sound, and video at the same time, with support for multiple languages including Chinese and English, as well as dialects and accents.

Writing requirements for multilingual scenes:

• Clearly write the speaker, language, lines, and tone item by item

• The same scene can include dialogue in different languages

• When the voice direction needs to be bound to a character, use character references together with sound control so the speaker and tone in the scene do not drift

• Keep the speaking face visible and the lines clear so lip movement and expressions remain natural and coherent

NadouPro’s audio models can additionally generate voiceover and music, which can be mixed with dialogue generated through audio-video sync.

Multi-Shot Storyboard Planning

Multi-shot generation supports up to fifteen seconds in a single generation, leaving room for dialogue, shot changes, and plot development. Taking a typical scene as an example, a cinematic sequence can be organized like this:

• Shot one: Full Shot, a European villa terrace, a man and a woman sitting

• Shot two: Close-Up of the woman as she talks about the leaves turning yellow

• Shot three: Close-Up of the man as he replies quietly

• Shot four: Medium Shot, the two in the same frame, the woman smiles

The cinematic language used should be visible in the output: shot-reverse-shot dialogue, cross-cutting, voiceover, Dolly In, Close-Up, Full Shot, and tracking shot.

Planning checklist:

• Reference consistency: choose clean reference assets and define each asset’s role in the scene

• Sound binding: pair speaker, lines, language, and tone

• Shot planning: define duration, shot size, angle, narrative content, and camera movement one by one

• Long scenes: plan story beats around the available duration

Camera movement can be described in natural language: a slow Dolly In, handheld tracking, a Full Shot establishing frame, or a low-angle tracking shot beside the subject.

Education and Science Visualization Scenarios

Educational content can use reference assets to keep the explanation subject consistent, then use prompts to describe visible steps, labels, motion, and scene changes.

• On-camera explanation: use a character reference plus clear voice direction so the speaker, lines, and visual role remain easy to understand

• Product lessons: use product or object references and describe what changes and what remains recognizable

• Step-by-step concepts: use multi-shot planning to place setup, demonstration, and result into different shots

• Frames with labels: keep labels short, clearly positioned, and part of the composition

In NadouPro’s canvas workflow, subtitles, diagrams, and demonstration clips for explainer videos can be generated as separate nodes and then stitched together, so changing one step does not require remaking the whole video.

Frequently Asked Questions

Q: How should reference assets be used in video prompts?

First associate the reference assets with the scene on the canvas or in the Video node, then describe in the prompt which character, which image, which video reference, and which type of sound is guiding the result. Reference assets define "what it is," while the prompt defines "how it happens."

Q: How should real-world movement be described?

Use visible motion language: speed, direction, weight, posture, contact with the environment, and camera movement. Do not write abstract phrases like "strong sense of motion"; make it clear which object is moving, how it moves, and in which direction.

Q: Which languages and audio capabilities does audio-video sync support?

It supports synchronized generation of dialogue, sound effects, and ambient sound, covering multiple languages such as Chinese and English, as well as dialects and accents. Multi-character dialogue should use clear speaker labels and tone descriptions.

Q: How does multi-shot generation work?

Multi-shot mode supports more shots and scene coverage within a single generation; when shot-level control is needed, specify duration, shot size, angle, narrative content, and camera movement for each shot, and the model follows the configuration strictly. NadouPro’s Storyboard and Clip can organize multiple shots into a complete finished video.

To explore more features, visit NadouPro: NadouPro

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