
This article uses a real-world case study of creating a promotional video for a coffee shop to walk you through how to leverage the latest AI video generation tools (such as Sora2 and Lovart) to craft...
Want to shoot a promotional video that makes your product “look expensive,” but your budget is only in the triple digits? Don't worry—it's no longer a pipe dream. I'm going to walk you through a real-world case study, teaching you step-by-step how to use the latest AI video generation tools to create a viral commercial for your product from scratch. You'll master 9 practical techniques from beginner to expert level, transforming AI from a “fun toy” into your “high-efficiency productivity tool.” Ready to revolutionize your understanding of video production?
Just before National Day, OpenAI launched its S2 model and app, enabling highly realistic video generation through simple facial and voice capture.
It's unsettling to think about—could the blurred lines of authenticity in AI-generated content usher us into the next era of information chaos?
Speaking of which, yesterday as the owner of an independent coffee shop, I was struggling with how to promote our upcoming autumn special, the “Osmanthus Latte.” Suddenly, I noticed my go-to AI tool Lovart had also rolled out S2 functionality.
This signaled S2's entry into industrial-scale production lines, poised to unlock immense commercial value.
As a budget-conscious entrepreneur, my priority was clear: Beyond novelty, how could I “implement” this into my workflow to cut costs and boost efficiency?
So I set myself a challenge: without assembling a team or outsourcing, I'd use only AI to create a “looks-expensive” commercial ad for my “Osmanthus Latte” from scratch.
The process involved hitting many roadblocks, but I distilled a set of practical techniques that let even “non-designers” create polished ads effortlessly.
Today, I'm sharing all 9 of these techniques with you.
At first, I was like a kid with a new toy—naively enthusiastic. I simply pointed at a product image of “Osmanthus Latte” and tossed a plain-spoken request at the AI:
Generate a brand advertisement video for this latte, in a cozy autumn style, featuring a beautiful Chinese model enjoying the coffee.The result? The AI delivered a video with genuinely impressive effects... but how to put it? It felt more like a high-end animated PowerPoint slide—utterly lacking the “premium feel” and “narrative depth” expected of an advertisement.
You've probably fallen into this trap before, right? You envision a blockbuster movie in your head, but the AI spits out something completely off the mark.
That's when I realized I needed to treat it like an “employee.” You have to be the director, giving it clearer instructions.
So I began my real exploration.
I ditched vague descriptions and started dissecting each shot into specific steps.
For example, instead of saying “beautiful woman drinking coffee,” I'd specify:
A woman in a beige sweater walks to the window, picks up a latte from the table, takes a gentle sip, and then looks out the window with a smile.S2 instantly “got it,” with the characters' movements, rhythm, and pauses in the video all perfectly on point. This is the first step in transforming a video from a ‘GIF’ into a “short film.”
The movements were right, but the texture still felt off. A professional commercial requires lighting, camera angles, and atmosphere to be spot-on.
So I tried a more professional format, clearly outlining all requirements like a call sheet for a film crew:
Shot on Arri Alexa, 35mm lens, gentle dolly-in, soft morning light filtering through a cafe window, steam rising from the latte. Actions: A character's hand enters the frame, stirring the latte gently with a silver spoon. Diegetic sound of spoon clinking and soft cafe ambiance.This time, the lighting, camera movement, and atmosphere came together instantly. See, S2 has an exceptional grasp of professional terminology—the more specialized you are, the more stunning the results it delivers.
Want to take it further and turn your video into a short drama? I decided to use this template to create a scene showcasing the “art of pour-over coffee”:
Style: Moody and cinematic with high contrast, inspired by Japanese minimalist aesthetics. The scene is filled with deep shadows and warm, focused light.``Inside a quiet, minimalist coffee bar, a single spotlight illuminates a wooden counter. On the counter sits a ceramic V60 dripper, a gooseneck kettle, and a single coffee cup. A barista, visible only by his hands and forearms in a crisp white shirt, performs the ritual of making pour-over coffee.``Cinematography:` `Camera: Medium close-up, fixed, eye-level.` `Lens: 50mm prime lens, shallow depth of field.` `Lighting: A single, warm spotlight from above, creating dramatic shadows.` `Mood: Calm, focused, almost meditative.``Actions:` `- The barista's hands slowly pour hot water from the kettle over the coffee grounds in a circular motion.` `- Steam rises elegantly from the dripper.` `- A single drop of coffee falls into the cup below in slow motion.``Background Sound:` `The gentle sound of water pouring, a faint, distant sound of a clock ticking.As you can see, this video already has a storyline and artistic flair compared to the previous ones.
The biggest fear when shooting commercials is running out of creative ideas and lacking a sense of cinematography. But I'm not formally trained—terms like “push/pull/pan/tilt” and “shot composition changes” mean nothing to me.
So I decided to “fight fire with fire.” First, I used Lovart to generate a mouthwatering ‘tiramisu’ image as my “opening shot.” Then I had Lovart Agent generate the storyboard script for me:
Using this image as the opening shot, generate a multi-angle shooting script for a dessert video. The script should have 8 coherent shots, structured by: Shot Type, Camera Movement, and Scene Description. Keep the language concise, under 500 characters.Once you have this script, you can use it directly to generate the video:
The effect is truly impressive—the lens conveys a strong sense of narrative and rhythm.
AI-generated scripts solve the “creativity” problem, but achieving pixel-level precision control often proves challenging—AI tends to be uncooperative.
This is where the “secret weapon” comes in—JSON-formatted prompts. Think of it as a “shooting checklist” for the AI director. I used this approach to generate a viral “Lego Castle Auto-Assembly” video:
{ "description": "A cinematic time-lapse of a LEGO castle building itself on a wooden table. The scene starts with a pile of LEGO bricks that quickly and magically assemble into a complete, intricate castle. The final shot is a majestic view of the finished castle.", "style": "magical realism, cinematic time-lapse", "camera": "starts with a top-down view, then orbits around the castle as it builds, ending with a low-angle dramatic shot", "lighting": "starts with bright, even light, transitioning to dramatic sunset lighting on the finished castle", "environment": "a cozy, slightly dusty workshop", "elements": [ "LEGO bricks (various colors and shapes)", "wooden table", "finished LEGO castle with towers and flags" ], "motion": "bricks fly and click into place rapidly and precisely", "ending": "a majestic, fully assembled LEGO castle glowing in the sunset", "keywords": [ "16:9", "LEGO", "castle", "stop-motion feel", "magical assembly" ] }Honestly, I hit a wall generating this JSON with S2—the results were mediocre. Just as I was about to give up, it hit me: what I should care about is “solving the problem,” not “evaluating the tool.” I promptly switched to Veo3—also from Lovart—and fed it the exact same JSON prompt. The results were instantly top-notch! This experience made me realize that the core of future AI applications lies in multi-model collaborative workflows.
With satisfying storyboard footage in hand, the next step is to seamlessly stitch them together into a complete commercial.
To give the commercial a more “cinematic feel,” I employed seamless transitions using a single continuous shot. First, I used AI to generate two images—one distant (city nightscape skyline) as the ‘opening’ and one close-up (latte by the café window) as the “closing.” Then, using Vidu Q2's keyframe feature, I input the prompt:
The camera flies at high speed over the glittering city skyline at night, seamlessly passing through the windows of a building and ending with a close-up shot of a latte on a table inside a cafe.The resulting video transitions are incredibly smooth, giving it a truly cinematic feel.
Sometimes my requests are genuinely complex—like generating an epic “ancient Roman battlefield” scene. The prompt can get extremely lengthy. That's where Lovart's Agent shines: dump all your messy requirements on it, and it'll independently think, plan, break down the task, then delegate execution to the underlying video model.

"The Last Centurion"``[Visual Style]``The video is presented in ultra-realistic, gritty photorealism with a cinematic 21:9 aspect ratio. Inspired by films like "Gladiator" and "Saving Private Ryan," the aesthetic is raw and visceral. The air is thick with dust, smoke, and dirt particles, rendered with high fidelity. The color palette is desaturated, dominated by earthy tones, mud, and the crimson of Roman cloaks, with glints of sunlight off steel. The image is sharp, with a deep depth of field to convey the vastness of the scene.``[Lighting]``The scene is set during the "golden hour" of late afternoon. The low-angle sun casts long, dramatic shadows across the battlefield. The light is harsh, creating high contrast and highlighting the texture of armor and terrain. Flickering light from distant fires and torches provides a secondary, chaotic light source.``[Camera]``The shot begins with a tight close-up on a Roman Centurion's grim, determined face. The camera then executes a slow, continuous pull-back, revealing his position on a hill, and finally, the massive, chaotic battle unfolding in the valley below. The movement is smooth and epic, like a large-scale crane or drone shot, emphasizing the scale of the conflict.``[Subject / Character]``The main character is a veteran Roman Centurion, standing alone on a small hill. He wears detailed Lorica Segmentata armor, visibly scratched, dented, and stained from battle. His red cloak (sagum) is tattered and whips in the wind. He holds a gladius, its blade clean, and his large scutum (shield) rests on the ground beside him, its face bearing the emblem of his legion. His face is weathered, with a prominent scar over one eye. He is stoic, observing the battle with the focused gaze of a commander.``[Environment]``A vast, muddy valley serves as the battlefield. Thousands of soldiers are locked in brutal combat. Well-drilled Roman formations are visible, a wall of shields against a seemingly endless barbarian horde. In the distance, smoke rises from burning siege engines and sacked villages. The ground is littered with broken weapons, discarded shields, and the fallen from both sides.``[Sound Design]``The audio is immersive and overwhelming. From the Centurion's vantage point, the roar of the battle is a constant, slightly muffled cacophony: the distant clash of thousands of swords, the blare of war horns, the rhythmic thud of legionary drums, and the collective shouts and screams of men. The primary sound is the lonely howl of the wind on the hilltop, whipping past the Centurion's helmet.``[Actions]``The Centurion remains almost perfectly still, a statue of resolve amidst the chaos. His only movements are the slight turn of his head as he surveys the battlefield and the slow, deliberate tightening of his grip on his gladius hilt. A gust of wind makes his cloak billow dramatically.``[Special Effects]``Atmospheric effects are paramount: volumetric smoke and dust clouds drift across the scene. Subtle motion blur is applied to the distant, fast-moving combatants to enhance the sense of chaos. A realistic lens flare appears as the camera pans past the setting sun. All effects are grounded in gritty realism.Armed with all the techniques above, I could finally return to my original goal—creating a commercial ad for my “Osmanthus Latte.” I combined the desired scenes, models, styles, and camera movements to generate the final ad footage. The entire process took only minutes, with computational costs amounting to just a few cents.
In the past, creating a cinematic-quality ad required a team, a budget of hundreds of thousands, and a lengthy production cycle. Now, all it takes is a spark of inspiration in your mind and a few lines of prompts.
Platforms like S2 and Lovart have truly placed the power of filmmaking in the hands of every ordinary creator.
Now, it's your turn.
If you had access to such AI video tools, what kind of video would you most want to create for your ________ (product/hobby/life moment)? Share your ideas in the comments!
Runway Gen-3 Alpha brings revolutionary AI video generation capabilities. This article conducts in-depth testing of its core functions, such as text-to-video and image-to-video, and analyzes its appli...
Shares AI Video, Video generation
This article details how to build a fully automated tech intelligence system using n8n automation, Docker local deployment, Zhipu AI for analysis, and Feishu/Lark bots for notifications. The tutorial ...
More in this section
Comprehensive guide to Claude Skills feature - learn how to transform AI from a one-time conversation tool into a sustainable professional collaborator through workflow engineering, complete with real...
More in this section
This article provides an in-depth review of Google's latest Gemini 2.5 Flash Image model. Through over 30 real-world test cases, it demonstrates the model's revolutionary ability to solve the subject ...
More in this section
Loading comments…