How to Write AI Video Prompts? NadouPro Canvas Workflow Guide: Practical Templates for Characters, Scenes, and Action

In AI video creation, "the generated result does not match expectations" is the most common pain point. The NadouPro Canvas Workflow uses three layers of control—character building, scene setup, and camera language—to let creators lock in key elements before clicking generate. This article organizes a set of verified prompt templates covering typical categories such as fight sequences, building destruction, multi-character scenes, and motion control.

Characters First: Build an Anchor for Consistency

Every film or video project starts with characters. In the Canvas Workflow, first use image fusion mode to generate character design sheets—front, side, and back three-view references—to ensure the character remains consistent in every later shot. Image fusion mode extracts facial features, skin tone, and visual style from multiple reference images, locking the character’s identity throughout the video.

When setting a character, the budget and era options in the cinematography parameters affect the overall visual tone. For example, if you set "action genre, 2010s," the system will automatically match the color grain and lens texture of that period.

Scene Generation: Use Spatial Depth Instead of a Flat Background

When generating a scene, one keyword is enough. For example, entering "hotel corridor" lets the Canvas Workflow automatically build depth, lighting, and atmosphere—not a flat corridor with a centered camera position, but a space with depth of field and layered lighting.

One tip: describing the scene from a three-quarter angle shows more spatial information than a front view, and the system understands depth more accurately.

Practical Prompt Templates

The following prompts use typical scenes as examples. Creators can adjust them according to their own projects.

Multi-shot fight sequence

Take a hotel corridor fight as an example. Upload a scene keyframe as the starting frame, and include the character design sheet and scene reference image to ensure consistency across shots.

Multi-shot edit, hotel corridor fight sequence. Shot 1: telephoto lens. A young man in a dark suit stands at the end of the corridor. The corridor has a deep red carpet, beige walls, and warm yellow wall lamps. Five or six men in black block the way, unbuttoning their jackets and taking fighting stances. Shot 2: handheld tracking shot. The protagonist strides toward the first opponent and drives him into the wall with an elbow strike in the narrow space; a picture frame shakes loose, falls, and shatters. Shot 3: low-angle side shot. The second man punches from the side; the protagonist grabs his arm and uses the force of the wall to throw him onto the carpet. Shot 4: fast frontal close-up. The protagonist dodges a punch, strikes the opponent’s chin with his palm, and the opponent crashes through a room door and falls. Shot 5: handheld shaky tracking shot. Two men close in at the same time; the protagonist blocks left and right, knees one man in the abdomen, and elbows the other in the temple. Shot 6: dynamic tracking shot. The final opponent guards the end of the corridor; the protagonist accelerates into a sprint, turns sideways to evade a heavy punch, spins back, and knocks him down with an elbow strike. Shot 7: fixed camera, medium shot pulling out to a wide shot. The protagonist calmly walks past the fallen crowd, straightens his tie with one hand, and stops in front of the last door.

Scene continuation without creating a new keyframe

Upload the previous video directly as context and write the next beat. The same face and the same outfit connect seamlessly from shot to shot.

Building demolition (text-to-video only)

No starting frame or reference image is needed; the video can be generated from the prompt alone. This tests how the Canvas Workflow handles physical effects and environmental realism.

A demolition crew detonates a skyscraper in a city at dawn. The building folds inward floor by floor, glass and concrete dust pouring down like a waterfall, a cloud of debris slowly spreading outward, and pigeons scattering into the orange sky. At street level, a handheld camera films from behind a safety barrier, with crowd reactions visible in the foreground. A realistic shockwave of dust rolls toward the camera.

Multi-character scene (materials only, no starting frame)

Upload 3 characters and 1 scene, and the Canvas Workflow automatically arranges the lighting while keeping each character’s face consistent across shots. This is director mode: you choose the cast and the setting, and the model makes the creative decisions.

Three men sit on folding chairs in front of a sun-baked concrete wall, under the blazing sun, with heat waves rising. One man has sweat on his forehead, squints, and casually says: "It sure is hot today." Close-up of another man: "It really is." Then a wide shot, the third man: "That’s nothing. You haven’t been to Almaty." Finally cut to a close-up of a barking dog.

Motion control (replacing the background of an existing video)

Upload a video of a person in motion and transfer that motion into a completely new environment. The Canvas Workflow automatically relights the scene and perfectly copies the motion path, with no manual masking or compositing needed.

In this video, replace the scene with reference image 1. Horror film style, a man is running away from something terrifying.

Product Ads and Motion Design

In addition to narrative content, the Canvas Workflow is also well suited to commercial scenarios. Types such as pizza ads and Logo animations can be driven directly through prompts. The key to product advertising is to describe the product material, lighting, and camera movement while excluding interference from people and text.

The product is placed in a minimalist kitchen on a dark stone countertop. Warm side light comes from a single window. The camera slowly pushes from a medium shot to a close-up, clean and premium. No people, no text, no Logo.

Frequently Asked Questions

Q: How long are videos generated by the Canvas Workflow?

Different video models support different durations. Seedance 2.0 supports multi-shot sequences up to 15 seconds, 可灵 3.0 Omni 4K supports multiple shots generated at the same time, and 3D 导演台 supports video export up to 90 seconds.

Q: How can I keep a character consistent across multiple shots?

Use image fusion mode to generate a character design sheet (three-view references), and upload it as reference material when generating the video. Image fusion mode locks facial features, skin tone, and visual style to ensure the character remains unchanged throughout the video.

Q: Should prompts be written in Chinese or English?

NadouPro supports Chinese prompts. It is recommended to describe scenes and actions in Chinese, while cinematography terms can remain in English (such as dolly in and tracking shot), and the system will recognize them correctly.

Q: What techniques help with choreographing fight scenes?

Number each shot separately and describe the specific action shot by shot. Use numbering such as "Shot 1" and "Shot 2" to clearly mark the camera position and action for each shot. Add environmental reactions to the action description, such as a picture frame shaking loose or a wall lamp shattering, to enhance realism.

Q: Are more detailed prompts better for scene generation?

At the scene generation stage, concise prompts are often more effective. A single keyword such as "hotel corridor" or "modern living room" can trigger the system to automatically build a complete space. Too much detail may instead limit the system’s creative output. The key is to add specific instructions only at the character and video stages.

Q: How can I modify the scene based on an existing video?

Use the motion control feature: upload the original video and a new scene reference image, then describe the change in the prompt. The system keeps the character’s motion path, replaces only the background environment, and automatically matches the lighting.

Explore more features on NadouPro: NadouPro

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