Vidu Video Generation Prompt Guide-Alibaba Cloud Model Studio(Model Studio)-阿里云帮助中心

A comprehensive guide to writing Vidu prompts, covering prompt formula, keyword reference, and advanced examples.

Getting Started

Prompt Formula

Subject/Scene + Scene Description + Environment Description + Art Style/Medium

Adjust sentence structure and word order to avoid having too many/complex subjects or scattered subject placement.
Avoid ambiguous terms; be as specific as possible.
Use fluent, natural phrasing and avoid overly literary descriptions.
Rich, accurate, and complete descriptions are essential for generating videos that match your desired art style and requirements.

Examples

Following the basic structure of "Subject/Scene + Scene Description + Environment Description + Art Style/Medium" improves video generation quality. Repeating and reinforcing key style descriptions across different parts of the prompt helps maintain consistent output style and enriches the visual result.

Using a "Corgi swimming" prompt as an example, the expanded version focuses on creating a warm, innocent cinematic visual effect.

Expanded prompt analysis: Start with the subject + scene + action, then layer in lighting, color tone, texture, and camera feel. Use aligned atmospheric words (gentle, soft, elegant, serene) at the beginning, middle, and end without mechanical repetition. The overall direction is warm, innocent, and cinematic.

Prompt: A corgi puppy swimming in a large pool

Expanded prompt (following the formula): Capture a serene moment of a corgi pup gracefully swimming in a sunlit pool. The underwater perspective reveals the puppy's gentle smile as soft golden light filters through the water surface, creating dancing shadows on the pool floor. The scene is rendered in soft pink tones, creating a dreamlike ethereal atmosphere. High-resolution cinematography captures every delicate detail of the water texture and the corgi's joyful expression, creating a simple yet cinematic portrait of tranquility and innocence. This minimalist, emotionally resonant setting conveys a sense of calm and happiness — an ideal cinematic shot that is both visually striking and deeply peaceful.

Prompt Keyword Reference

Specific keywords trigger different generation effects in Vidu. For example, adding "high motion" to the prompt increases the movement intensity in the generated video.

Category	Subcategory	Supported Interfaces	Keywords
Subject Motion	Motion Intensity	Text-to-Video Image-to-Video Reference-to-Video	High motion: high motion, fast movement, extra-large motion
			Medium motion: medium motion, moderate motion effect
			Low motion: subtle motion, slight movement
Camera Movement	Basic Camera Movement	Text-to-Video Image-to-Video Reference-to-Video	Pan left: camera pans left, camera move left, left pan, camera track left, etc.
			Pan right: camera pans right, camera move right, right pan, camera track right, etc.
			Tilt up: camera tilts up, camera move up, upward tilt, etc.
			Tilt down: camera tilts down, camera move down, downward tilt, etc.
			Zoom in: camera zooms in, camera zoom in, zoom-in shot, etc.
			Zoom out: camera zooms out, camera zoom out, zoom-out shot, etc.
			Static: camera stays still, static shot, fixed camera, etc.
	Special Shooting Techniques	Text-to-Video Image-to-Video Reference-to-Video	Time-lapse: time-lapse photography, time-lapse shot
			Macro: macro lens, macro shot
			First-person: first-person perspective, FPV shot
			Aerial: aerial shot, drone shot
	Shot Size	Text-to-Video Image-to-Video Reference-to-Video	Long shot
			Full shot
			Medium shot
			Close-up
			Extreme close-up
	Camera Angle	Text-to-Video Image-to-Video Reference-to-Video	Low angle
			High angle
			Eye-level
			Bird's-eye view
	Composition	Text-to-Video Image-to-Video Reference-to-Video	Diagonal composition
			Rule of thirds
			Symmetrical composition
			Over-the-shoulder: over-the-shoulder shot, over shoulder shot
			Dutch angle: Dutch angle shot, tilted frame
Video Style	Visual Style	Text-to-Video Reference-to-Video	2D anime: 2D anime style, anime illustration
			3D render: 3D rendering style, 3D CG style
			Realistic: realistic
			Ink wash: ink wash painting style, traditional ink style
			Line art: line art style, outline illustration
			Pixel art: pixel art style, retro pixel
			Oil painting: oil painting style, painterly
	Director Style Anime only	Text-to-Video	Hayao Miyazaki style
			Makoto Shinkai style
			Katsuhiro Otomo style
			Mamoru Hosoda style
			Goro Miyazaki style
			Masaaki Yuasa style
			Shin'ichiro Ushijima style
			Naoko Yamada style
			Takeshi Ishidate style
			Isao Takahata style, Yoshiaki Nishimura style
Visual Effects	Effects	Text-to-Video Image-to-Video Reference-to-Video	Explosion: explosion effect, explodes
			Vortex: vortex effect, swirling
			Shatter: shatter effect, shatters
			Warp: warp effect, warps
			Melt: melt effect, melts
			Disappear: disappears into bubbles, dissolves into snowflakes, etc.
			Petrify: turns into stone
Atmosphere	Atmosphere	Text-to-Video Image-to-Video Reference-to-Video	Joyful: bright colors Melancholy: cool tones Serene: soft tones Tense: dark lighting Warm and romantic: warm tones, soft lighting

Prompt Optimization

First draft prompts often don't yield ideal results. Identifying issues and iteratively refining the prompt is an essential optimization step.

Image-to-Video

Optimization analysis: The original prompt sometimes produces a visual effect where the train appears to move backward. The optimized version uses the size ratio between the subject and background to create a clear sense of forward motion.

Intended Effect

Input Image

Original Prompt

Original Video Result

Optimized Prompt

Optimized Video Result

Train moving forward

The train starts moving forward

# Subject The train starts moving forward, the front of the train gradually gets larger, smoke continuously billows from the chimney, and pine trees shrink into the distance.

# Requirements:

1. Camera movement: follow to the right

Reference-to-Video

Optimization analysis: The walking motion in the original prompt looks unnatural, possibly due to a static camera. The optimized version adds camera control (zoom out), which makes the character's movements look more natural when combined with camera motion. If you need the prompt to work with various input images, continue adjusting using the techniques described above.

Original prompt example:

Intended Effect

Input Image

Prompt and Result

Nature-style dynamic portrait - solo

Original prompt: The woman walks lightly toward the camera, her steps firm and graceful. On a forest path, sunlight filters through dappled tree shadows onto her shoulders. A gentle breeze stirs her skirt as leaves flutter softly with each step, conveying a sense of ease and composure.

Optimized prompt

# Subject The woman walks lightly toward the camera, her steps firm and graceful. On a forest path, sunlight filters through dappled tree shadows onto her shoulders. A gentle breeze stirs her skirt as leaves flutter softly with each step, conveying a sense of ease and composure.

# Requirements

1. Camera movement: zoom out

Advanced Guide

Maintaining Subject Consistency in Reference-to-Video

Vidu maintains subject consistency across multiple perspectives, supporting complex character designs, facial expressions, and multi-subject interaction scenarios.

Example 1: Precise Control of Complex Subjects

For complex 3D animated characters, Vidu consistently preserves head shape, costume details, and full-angle consistency regardless of viewpoint changes.

Example 2: Natural Facial Feature and Expression Consistency

In close-up portraits, Vidu naturally and smoothly maintains facial feature and expression continuity, avoiding stiffness or distortion. Fine facial control gives Vidu a significant advantage for creating nuanced, realistic characters.

Example 3: Multi-Subject Consistency

Vidu supports uploading multiple subject images (characters, props, environments, etc.) to create interactive scenes. For example, uploading images of a character, an object, and an environment lets you generate a scene with a custom character performing actions in a custom space.

Vidu also supports fusing features from different subjects — for example, seamlessly combining the front of character A with the back of character B to create an entirely new character or object.

AI Manga Drama Prompt Examples

The prompt writing principles are the same as the getting started guide (avoid too many/complex subjects, avoid ambiguous terms, use natural phrasing, keep descriptions rich and accurate).

Manga drama prompt structure:

Part 1. Style / Shot Size / Camera Position / Composition / Camera Movement (optional)

Style: 2D manga, realistic, 3D anime, Korean webtoon, Lolita, otome game, game concept art, medieval, ancient Chinese, martial arts, cartoon, picture book, blind-box collectible, cyberpunk, tech, metallic, dark horror, etc.

Shot Size: Long shot, full shot, medium shot, close-up, extreme close-up

Camera Position: Left of axis, right of axis, low angle, high angle, looking up, looking down, eye-level, bird's-eye, drone, etc.

Composition: Rule of thirds, horizon line, center, Dutch angle, etc.

Camera Movement: Static, up/down/left/right, diagonals, zoom in, pan/tilt, zoom out, clockwise/counterclockwise orbit, rotation. Movements can be combined (e.g., "zoom in + clockwise orbit").
Part 2. Scene Description

Describe the action, scene, or visual effects in simple natural language.
Part 3. Image Reference

Image 1 is [scene]; Image 2 is [character xx]; Image 3 is [prop xx]

Prompt Examples

Prompt Example	Reference Images	Result
2D anime style, medium shot, action freeze, static camera; horizon line composition, right of axis, Xiao Lan, Xiao Hong, and Xiao Ming are walking through the peach blossom valley (Image 1), smiling, when suddenly their smiles freeze. Image 1 is [scene]; Image 2 is character [Xiao Lan]; Image 3 is character [Xiao Hong]; Image 4 is character [Xiao Ming].	Images 1-4 in sequence:
2D anime style, camera moves from bottom to top, avoiding the face — starts with a close-up of footsteps stepping forward, then slowly pans up to the chest without revealing the face; static camera, in Image 1, a massive beam of light slowly reveals the true form of the evil old man as he walks out, radiating an eerie and sinister energy. Image 1 is [scary, eerie, awe-inspiring environment]; Image 2 is [evil old man].	Images 1-2 in sequence:
3D anime style, close-up, eye-level, left of axis, rule of thirds, Xiao Fei is lifted into the air by a swirling vortex, his hands desperately gripping a rock jutting from the ground. In the fierce wind, Xiao Fei struggles to look back and shouts behind him, his expression fierce, mouth constantly moving, the camera shaking violently. Image 1 is the scene; Image 2 is character [Xiao Fei].