3-Step Workflow To Make Ultra-Realistic AI Ads

How to Make Ultra-Realistic AI Ads

This video demonstrates a three-step workflow for creating a complete cinematic commercial with AI. The process begins with building consistent assets, continues with AI-assisted shot-list and prompt creation, and finishes with iterative scene generation and editing.

Step 1: Build the Assets

The first stage focuses on creating and organizing all the visual assets needed for the commercial, including the product, hero character, supporting characters, locations, and props.

Create a Product Reference Sheet

A single product image may not provide enough visual information for consistent video generation. The workflow creates a product sheet showing multiple angles, including front and three-quarter views, so the video model has better references to work from.

Create Consistent Character Sheets

The main character appears throughout the commercial, so a detailed character sheet is created with close-up and full-body views. The close-up establishes the face, while the full-body view provides information about height and build.

Multiple character candidates are tested rather than choosing a character based only on a still image. The preferred character is selected after testing how well the face and appearance hold up during video generation.

Build Supporting Characters

Secondary characters do not require the same level of testing as the hero. A character-generation tool can quickly produce character sheets and styling variations for supporting roles such as the office boss.

Create Cinematic Locations

Locations are generated in multiple variations and tested before being locked. The workflow favors bright, clean, high-budget commercial aesthetics and often uses three-quarter views to provide greater depth for camera movement.

The kitchen, stadium, street corner, and office are developed separately. Each location is compared through video tests before the strongest version is selected.

Test One Variable at a Time

The same prompt is reused while changing only the character or location. This makes it easier to understand which asset produces the better result instead of changing multiple variables at once.

Prepare Assets for Consistency

When a character sheet contains multiple faces, the workflow removes the unnecessary face from the full-body reference so the video model has a single facial reference to follow.

Locations can also be edited before animation. For example, the kitchen is modified to create space for the coffee-making action and other story requirements.

Create Multiple Versions for Character Changes

When a character needs to change appearance during a scene, separate references can be created. For the stadium sequence, the workflow creates both a dry athletic version and a post-run wet version of the hero character.

Preserve Image Quality

Repeated AI editing can reduce image quality and create a flat or artificial appearance. The workflow combines the original high-detail character image with the edited clothing version in a photo editor, preserving the original face, skin, background, and overall detail while replacing only the outfit.

Create Prop References

Recurring objects such as sneakers, backpacks, coffee equipment, and other props receive their own reference sheets. This helps maintain consistency between shots.

Step 2: Create the Shot List and Prompts

The second stage turns the script and locked assets into a connected shot list rather than a collection of unrelated prompts.

Use a Claude Skill for Shot Lists

The workflow uses a custom skill in Claude that is designed to generate shot lists and prompts for the chosen AI video workflow. The skill contains instructions about shot structure, character consistency, and common problems encountered during video generation.

Give Claude the Script and Assets

The script, locked visual assets, and asset names are provided together. Instead of simply describing the assets, the workflow uploads the actual reference images so the AI can see exactly what each asset looks like.

Create a Connected Prompt System

The shot list contains a shared style prefix covering lighting, camera, color, and the overall visual treatment. Because the prefix is connected to every prompt, a global change can be applied across the entire commercial without rewriting every shot individually.

Each prompt also receives a unique name, allowing individual shots to be edited without affecting the rest of the shot list.

Step 3: Generate and Refine the Scenes

The third stage involves generating the scenes, identifying failures, and repeatedly refining the prompts. The creator emphasizes that the first generation rarely works perfectly.

Fix Lighting and Camera Movement

Early generations may have incorrect lighting, static cameras, or too many actions packed into one shot. Instead of forcing everything into a single generation, the workflow separates important moments into dedicated shots.

Break Complex Actions Into Separate Shots

The coffee-making sequence is separated from the main kitchen entrance. Dedicated references are created for the moka pot and mug, and the coffee process becomes a fast-cut montage with close-ups of the preparation, flame, pour, and sip.

Use Detailed Choreography

Generic instructions such as “dance” can produce inconsistent movement. The workflow instead describes individual movements such as head nods, shoulder rolls, knee dips, finger snaps, spins, footwork, and other specific actions.

Use Match Cuts

When scenes connect, the workflow matches the opening and closing actions between shots. For example, the same hand and motion are used to connect the end of one scene with the beginning of another.

Stadium Scene

The stadium sequence introduces an athletic version of the hero and uses separate dry and wet character references. Scene-specific lighting overrides are also applied when the visual requirements change from the shared commercial style.

The sequence combines broadcast-style wide shots, tracking shots, low-angle shots, and a dedicated product shot focused on the headphones.

Using Layout Maps for Spatial Consistency

One of the workflow's key techniques is creating a schematic layout map when text alone cannot reliably maintain the position and scale of objects.

The map identifies important elements such as a fire hydrant and sky dancer, including their relative position and size. The schematic is then provided to the prompt-writing workflow so the scene can maintain more consistent geography between generations.

Locking Character Position and Props

When the hero drifts between generations, a fixed environmental anchor such as a tree can be used to establish his position. Important props, such as a backpack, are explicitly locked in the prompt so they appear consistently across cuts.

Synchronizing Movement With Music

The actual music track can be supplied as an input and the character's choreography can be instructed to follow the beat. This helps make individual dance movements feel connected to the soundtrack.

Office Climax

The office sequence brings the hero and angry boss together. The entrance is refined by matching the previous scene and describing the hero's dance movements individually while the camera follows him through the office door.

Final Scene

The final scene reverses the earlier joke by showing the previously angry boss outside wearing headphones and dancing beside the sky dancer. The contrast provides the commercial's closing comedic beat.

Iteration Is the Core Skill

The finished commercial is assembled from the strongest moments across many generations rather than relying on one perfect generation. The workflow emphasizes testing, selecting the best clips, fixing individual problems, and stitching the strongest seconds together.

Complete 3-Step Workflow

1. Build the assets: Create and lock consistent products, characters, locations, and props.
2. Create the shot list: Use the script, reference assets, and a connected prompt system to generate structured scene prompts.
3. Generate and iterate: Create scenes, identify problems, refine prompts, generate alternatives, and edit together the strongest results.

Key Takeaways

1. Create multiple asset options before committing to a final version.
2. Test characters and locations in motion, not only as still images.
3. Use reference sheets for products, characters, and recurring props.
4. Change one variable at a time when testing.
5. Use connected shot lists instead of managing many unrelated prompts.
6. Use detailed choreography instead of generic movement descriptions.
7. Create layout maps when text prompts cannot reliably control spatial relationships.
8. Create separate reference versions when a character's appearance changes.
9. Use dedicated product shots to highlight the product.
10. Expect multiple generations and select the strongest moments during editing.

Conclusion

Ultra-realistic AI advertising does not depend on a single perfect prompt. The workflow combines careful asset preparation, structured prompting, visual references, spatial planning, motion testing, and repeated iteration. The same approach can be adapted to products, brands, commercials, and original creative concepts.