How to Make a Cinematic Sports Ad: NadouPro Workflow for Scene Prompts and Asset Locking

Behind a thirty-second cinematic ad is the repeated appearance of products, characters, scenes, and props across multiple shots. If you start generating video directly, the shots before and after will inevitably fail to match: the product gets different packaging, the character gets a different face, the scene gets different lighting, and the whole film falls apart.

The correct order is the opposite: first turn every recurring asset into locked reference images, then generate the video scene by scene. This tutorial breaks down the full workflow with a sci-fi ad in the style of a sports drink commercial, and all prompt structures can be reused directly.

Before Production: Build an Asset Library First

Every element that appears repeatedly in the ad needs a corresponding reference image. For this ad, six types of assets need to be prepared:

Product three-view sheet. One product image is not enough; the model needs to know what it looks like from every angle. Use Multi-angle generation to create front, back, and top views. The prompt is very simple:

Based on this product image, generate a product three-view sheet containing front, back, and top views.

Character reference sheet. The protagonist’s face and body should be locked into a reusable reference sheet, with a full-body portrait on the left and a facial close-up on the right:

Generate a character reference sheet with a two-column layout: full-body portrait in the left column and facial close-up in the right column.

You can deliberately avoid fixing the costume, let the model return multiple looks, and then choose from them.

Mecha character three-view sheet. The sci-fi mecha in the ad should also be made into a two-view character sheet, labeled front and back. Write the key details clearly: armor proportions, color distribution, chest number, and the positions of glowing components. Reuse the same descriptive language for the opposing mecha, changing only the number and color scheme, and the two characters will look like they belong to the same world.

Scene image. This is the most important image in the whole film—the video will inherit texture and lighting from this still frame. When describing it, add an anamorphic lens and film grain to suppress an overly clean digital look:

Modern city street, empty, sunny day, anamorphic lens, shallow depth of field, film grain.

Prop images. Balls, glowing props, and other items that appear in the film should all be generated as standard product photos on a solid-color background, making them easier to composite later.

Step Two: Establish a Unified Style Header

Every scene prompt begins with the same style description. This is the key to making the whole ad look like the same film. The style header includes:

• Aspect ratio and image quality: large format, 8K, real photographic texture, explicitly excluding a 3D-rendered look;

• Lighting: natural light only, backlight, atmospheric haze;

• Color: primary color, secondary color, and accent color allocated in a six-three-one ratio;

• Skin texture: pore-level realism, fine hair, asymmetrical moles, changes in blood tone;

• Performance: a slight pause before reacting, precise eyeline, catchlights in the eyes;

• Physical rules: respect gravity and inertia, with no floating props;

• Continuity: characters, props, and scenes remain completely consistent in all shots, with identity drift prohibited;

• Audio: use ambient sound effects only, with no music; leave music for post-production.

Write the style header once, reuse it for all scenes, and do not change a single word.

Step Three: Generate Scene by Scene

Each scene prompt is organized into a five-part structure:

1. Style header: the unified description from the previous step;

2. Subject: reference assets through Multi-reference input, and label the position of each asset in the frame;

3. Action: describe in sections by shot number, with timecode and shot size marked for each section;

4. Camera: handheld or aerial, Dutch Angle degree, and the Focal Length of each shot;

5. Constraints: ending method, prohibited elements, and details that must be maintained.

Taking the opening "floating prop" shot as an example, the action section can be written like this:

Shot one: the protagonist suddenly stops while walking. The handheld camera moves around to the side in a low-angle medium close-up, and he narrows his eyes while staring ahead. Shot two: an Over-the-Shoulder Shot reveals a glowing product floating in the center of the empty street, gently moving up and down and slowly rotating. Shot three: he slowly approaches, and white light falls on one side of his face, creating a cool-warm contrast with the warm sunlight. Shot four: he extends his right hand, and the product descends under control, landing steadily in his palm. Shot five: the moment they make contact, the light fades and it becomes an ordinary product; he gives a small smile of relief.

The slow-motion rhythm is also written into the prompt: slow down during the reveal section, slow down again during the approach, make the moment it lands in the hand the slowest, then return to normal speed afterward.

Step Four: Handling Transformation and Fight Scenes

Transformation shots are the challenge in this type of ad. The key is to write the process as physical action rather than magic: armor plates fly in from the air, strike with weight, and lock into place according to physical logic. This feels more like live-action footage than particle effects. Give each component lock-in its own Close-Up, and combine rapid switches between handheld and FPV drone perspectives to create rhythm.

Use a low angle in the standoff scene to build pressure: a leg Close-Up of the opponent charging in from a distance, a Low Angle shot of the protagonist lowering his head into a ready stance, and a tracking shot of the fight for possession. Mark the viewing height and Focal Length for every shot; the tension of the action scene depends entirely on camera placement design.

Step Five: Editing and Finishing

NadouPro supports collaborative canvases for up to one hundred people, so multi-shot projects can be arranged on the canvas: place the scene-by-scene generated clips in script order, use the First/Last Frame feature to fill in transition shots, and add voiceover and sound effects uniformly during editing. Music is deliberately left for post-production so the generation stage can focus on the authenticity of the visuals and sound effects.

FAQ

Q: Why not generate the full ad directly? Why make it scene by scene?
A: The length and complexity of a single video are limited. Generating scene by scene allows precise control over each shot; if one shot fails, you only need to redo that section. Asset references ensure continuity after the shots are assembled.

Q: How precise do the asset reference images need to be?
A: They only need to show all key details clearly. Products need three-view sheets, characters need a full-body view plus a facial close-up, and the scene image needs to contain the final film’s shot texture.

Q: What role does the scene image play in video generation?
A: It is a style reference, not a locked frame. The model inherits its lighting, architectural style, and atmosphere, but the space inside the shot can extend freely. Do not lock the composition too tightly.

Q: How do you control the shaky feeling of handheld shots?
A: Clearly write the shake intensity and rhythm in the camera section, such as "strong operator shake, continuous organic jitter, no stabilization." The model will follow the description.

Q: What if I need realistic sound effects?
A: List the ambient sounds one by one in the audio section: wind, footsteps, and metal impacts. During generation, use an audio-video synchronization model so the sound effects and visuals are produced in the same pass.

Explore more features on NadouPro: NadouPro

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