I Made a Viral AI Love Story (500M Views) — Steal My Prompts

How a Viral AI Love Story Was Created

A love story created for a friend's wedding reportedly reached around 500 million views across social media. This video breaks down the complete workflow used to create the film, from character and location generation to scene prompting, storyboarding, physics, audio references, and final editing.

Step 1: Organize the Assets

The workflow starts with the assets: characters, locations, props, and other elements. A dedicated project is created with folders for every scene and subfolders for individual assets. Organizing everything before generating footage helps manage large numbers of generations later.

Creating Consistent Character Sheets

The first challenge is creating realistic character sheets from real photographs. Simply using a person's photos as references can produce a lookalike rather than an accurate representation.

The workflow uses a hybrid approach. The generated character sheet provides the costume, body, silhouette, and clothing details, while the actual person's photograph is placed into the portrait area. This approach is presented as a way to achieve more natural-looking faces while maintaining consistent outfits.

Generating Locations

Locations are treated as a major part of the video's visual quality. For the opening 1900s ship-departure scene, several location variations are generated and evaluated for crowd space, running room, lighting, clutter, and camera movement.

The selected location provides a long pier for the action while keeping the center of the frame relatively clean. Multiple batches are generated before locking the final location.

Using AI to Create Video Prompts

Instead of manually writing every video prompt, the workflow uses an AI assistant to expand simple scene descriptions into detailed prompts containing director-style instructions, timing, physics, camera angles, and shot details.

Scene One: The First Meeting

The opening scene takes place in 1905 at a busy departure pier. The two characters do not know each other yet. They move in opposite directions through a crowd and accidentally collide.

The first generations interpret the collision too romantically, turning a simple shoulder bump into hugging or kissing. The solution is to describe the physical mechanics precisely: shoulders make contact, suitcases swing, momentum carries the characters slightly past each other, and their planted feet turn.

The background characters are also instructed to remain active instead of standing still and staring into space. These changes make the collision feel more natural and grounded in physics.

Scene Two: The Gladiator Fight

The second major scene places the hero in a gladiator arena against a giant opponent. The sequence includes a fight, a sword, a takedown, and the hero looking toward the queen's box for a reaction.

Solving Character Scale Problems

A major challenge is keeping the giant's size consistent. The giant could appear dramatically different in different generations. To solve this spatial problem, a simple height-comparison sketch is created showing the relative size of the hero and the giant.

The sketch is then used as a visual reference alongside the detailed prompt. This helps establish the intended character proportions and physical relationship.

Splitting Action and Acting

The gladiator scene is divided into two parts. The first focuses on physical action such as grabbing, throwing, dodging, and takedowns. The second focuses on acting, emotion, reactions, and the characters' expressions.

The reason for splitting the scene is that a single prompt may struggle to give equal attention to complex physical action and detailed facial performance. The separate generations can then be stitched together during editing.

Train Robbery Scene

The next major sequence is a western-style train robbery involving horseback action, a shootout, and the two characters continuing their romantic story within the action.

Because the scene contains complicated spatial relationships, the initial generations struggle with the positions of horses, train tracks, vehicles, and jumps.

Using Storyboards for Spatial Logic

The workflow applies the same principle used for the giant's height problem: when a problem involves spatial logic, draw it rather than relying only on text.

Simple hand-drawn storyboards are created for individual shots. These sketches are uploaded alongside the textual description so the AI can understand the intended geometry and movement.

The resulting generations better preserve the horse's position, trajectory, and relationship to the train and surrounding objects.

Pirate Battle

The story then moves into a pirate sequence featuring two captains, a duel, a ship deck, a spyglass, doors, and background crew members.

Choosing the Right Model for Each Asset

The character-sheet workflow is tested with multiple image-generation models. The video explains that different models can have different strengths, such as clothing texture, hair consistency, or facial detail. The recommended approach is to choose the model that performs best for the specific asset rather than relying on a single model for everything.

Creating Consistent Background Characters

Randomly generated background pirates can change appearance between shots. To solve this, multiple pirate characters are generated together and saved as a single crew element. This allows the same background characters to be reused throughout the scene.

Why AI Video Requires Multiple Generations

The creator explains that a perfect 30-second AI render is not always realistic to expect. Instead, raw footage is generated in separate pieces and the strongest shots are selected and assembled during editing.

Italian Carnival Scene

The story moves to old Italy at night, with fireworks, a large masked ball, and the two characters dressed in burgundy. The location is generated with the crowd already present because empty environments were causing problems when the crowd was added later.

The scene initially works as one longer generation, but the emotional moments are too compressed. It is therefore divided into three shorter prompts, giving each story beat more room to develop.

Breaking Complex Scenes Into Shorter Prompts

The carnival sequence demonstrates an important workflow principle: complex emotional sequences can benefit from being divided into smaller generations. Breaking the scene into separate 15-second sections allows the characters to move through the crowd and react more naturally.

The Final Present-Day Scene

The final scene returns to the present day in a college elevator. The video reveals that the historical adventures shown throughout the film represent what was playing in the hero's imagination when he was trying to build up the courage to talk to the woman he eventually met.

Using Audio References

The final scene demonstrates the importance of audio references. A friend records a simple voice memo humming the main theme. Without the audio reference, the AI generates different melodies between takes.

With the audio reference attached, the generated characters follow the intended melody more closely. The video presents audio references as an important technique for maintaining consistency in musical or performance-based scenes.

Complete AI Love Story Workflow

1. Organize the project into scenes and asset folders.
2. Create consistent character sheets.
3. Use real photographs for more natural character faces where appropriate.
4. Generate and carefully select locations.
5. Use AI to expand simple scene descriptions into detailed video prompts.
6. Describe physical movements and physics precisely.
7. Use sketches when spatial relationships are difficult to explain with text.
8. Split complex action and acting into separate generations.
9. Create reusable background-character elements.
10. Generate multiple takes and select the strongest footage.
11. Use shorter prompts when emotional beats feel rushed.
12. Use audio references when musical consistency matters.
13. Assemble the strongest generations during editing.

Key Takeaways

1. Asset organization becomes increasingly important as AI video projects grow.
2. Character consistency requires more than simply uploading reference photographs.
3. Locations should be selected for composition, movement, lighting, and available space.
4. Physical actions should be described in precise mechanical terms.
5. Visual sketches can solve spatial problems that text prompts struggle with.
6. Complex scenes often work better when divided into multiple generations.
7. Background characters can be created as reusable elements for consistency.
8. Multiple AI generations are normally required to create polished footage.
9. Audio references can help maintain musical and performance consistency.
10. The final film is assembled from the strongest individual AI-generated shots.