Most AI ad campaigns fall apart for the same reason: the assets do not hold together. The character looks different in every shot. The product changes shape between frames. The lighting is inconsistent. Social posts look one way and the cinematic hero ad looks another. The result reads as a collage, not a brand.
This guide walks through the complete production workflow for building a cinematic AI video ad campaign — from a single creative brief to finished motion assets. Using the fictional pizza delivery brand Slice Riot as a worked example, it covers how to lock brand identity before generating a single frame, how to build character and vehicle reference sheets that enforce visual consistency across every scene, and how to write AI image prompts for advertising that produce professional, on-brand results every time.
The article also explains the difference between polished hero scenes and authentic UGC content layers, shows how to edit existing frames without breaking character consistency, and covers which motion models to use — and when.
Every prompt used in the production is included in full. You can copy them, adapt them to your own brand, and apply the same techniques in any creative studio.
Who it's for: Creative directors, brand marketers, AI content creators, and anyone who wants to produce a cohesive AI commercial or video campaign without assets that drift from shot to shot.
What You Will Learn
By the end of this tutorial you will know how to:
- Launch an AI video ad campaign from a single creative brief
- Lock brand identity before generating any asset
- Build a character reference sheet that holds across every scene
- Build a vehicle or product reference sheet with the same discipline
- Compose cinematic hero scenes using reference images
- Edit and refine frames without breaking character consistency in AI image generation
- Generate a social and UGC content layer that feels authentic
- Add cinematic motion to still frames using the right models for each job
These techniques apply to any AI video ad campaign regardless of category. The Slice Riot example is concrete, but the workflow scales to product launches, brand films, regional rollouts, and reactive campaigns.
The Core Problem Nobody Talks About
Most AI commercial productions fall apart for the same reason, and it is not the tools.
The character looks confident and sharp in frame one. By frame six, the jawline has shifted. The lighting on the car is warm in the hero scene and cold in the UGC shots. The social posts look like they came from a completely different creative team than the cinematic opener. The brand does not read as a brand — it reads as a series of disconnected experiments.
This happens because people start generating scenes before they have established what the campaign actually looks like. They treat each shot as its own problem, and the model treats each prompt as its own problem too. The result is technically impressive on a frame-by-frame basis and incoherent as a whole.
The fix is not a better model. It is a better process.
👉 Create Professional Videos with AI without any complicated process with Zoviz’s AI Video Generator
The Core Principle: Lock Identity Early, Reference It Everywhere
A reference sheet is a single generated image — or a tightly composed set of images — that locks the visual identity of one recurring element. The character. The vehicle. The product. The key location.
Every subsequent scene is generated using those reference sheets as visual anchors. You pass them directly into the prompt. The model has something concrete to match against, so consistency is enforced rather than hoped for.
This is the single decision that separates campaigns that hold together from campaigns that drift.
Generate scene by scene without reference sheets and you will spend more time fixing inconsistencies than you spent generating shots. Build the reference sheets first and every downstream prompt becomes faster, not slower.
Step 1: Start With a Creative Brief
The Slice Riot project began with a single line: a campaign for an ultra-fast pizza delivery service.
That is enough. A good AI video ad campaign does not need a 40-page agency brief before a single asset is produced. It needs a clear concept, an emotional register, and one distinctive idea the campaign can build around. Everything else is decided as the work progresses.
What makes a strong starter brief:
- A product or service described in one sentence
- An emotional register: cinematic, playful, premium, gritty, internet-native
- One distinctive tension or idea the campaign can hang on
For Slice Riot, the distinctive idea was that fast delivery should feel cinematic — almost dangerous. That creative seed shaped everything downstream, from the character's expression to the UGC aesthetic to the motion choices.
👉 Connect the Zoviz MCP in Claude: https://zoviz.com/mcp
Step 2: Lock Brand Identity Before Generating Any Asset
Before character sheets, before hero scenes, before social posts, you decide:
- Typography system. Wordmark style, headline font, body font, UGC overlay font.
- Color palette. Primary, secondary, and accent colors with hex values.
- Tone of voice. How the brand speaks on a billboard versus a reply on X.
- Visual language. Photography style, illustration approach, motion style.
- Brand personality. Three to five adjectives that describe how the brand feels.
For Slice Riot, the identity landed on bold and cinematic paired with an internet-native social aesthetic. The brand could plausibly appear on a Times Square billboard and as a rough meme on X on the same day — a deliberate dual register that shaped every asset that followed.
Write these decisions down before you generate anything. Every prompt that skips this step has to invent the brand identity on the fly. No two prompts invent it the same way. That is how campaigns drift.
👉 Create Marketing Videos, Ads & Content in Minutes with Zoviz’s Marketing Studio
Step 3: Build a Character Reference Sheet
If your AI video ad campaign features a recurring character, this is the most important asset you will produce. Build it correctly and every later scene becomes faster. Build it carelessly and nothing downstream holds together.
Here is the exact prompt used to generate the Slice Riot driver character sheet.
Prompt: Create a cinematic character sheet of a male driver with a strong, calm, and confident presence.
Character:
- male, late 20s to early 30s
- well-defined facial structure
- natural, realistic features (not model-perfect)
- clean skin, healthy appearance
- light stubble (optional, very subtle)
Face Condition:
- no cuts
- no bruises
- no scars
- no marks or damage on the face
- skin must look natural and intact
Expression:
- calm and controlled
- focused
- confident under pressure
- minimal emotion, strong presence
Outfit:
- minimal, clean, practical clothing
- dark jacket or hoodie
- no logos, no branding
Lighting:
- cinematic, low-key lighting
- soft highlights on face edges
- subtle contrast, not harsh
- natural falloff into shadows
Style:
- cinematic, film-like
- slightly stylized but grounded in realism
- cinematic, film-like
- slightly stylized but grounded in realism
- premium look without being commercial
Character Sheet Layout:
- front portrait (neutral expression)
- 3/4 angle
- side profile
- close-up on eyes
expression variations:
- calm focus
- slight intensity
- neutral
Camera:
- 85mm portrait lens feel
- shallow depth of field
- clean framing
Background:
- neutral dark background
- no environment storytelling
Texture:
- subtle film grain
- slight softness (not overly sharp)
- natural skin texture (no beauty retouch)
Important:
- same identity across all views
- no stylization or exaggeration
- no text
- no UI
Why this approach works for character consistency in AI image generation
The neutral background strips everything except the character — which is exactly what a reference sheet needs. The explicit list of forbidden elements (model-perfect skin, beauty retouching, scars, cuts) blocks the most common unwanted AI insertions. The locked camera language — 85mm portrait lens, shallow depth of field — gives every variation the same visual treatment. And specifying the full layout in one prompt gives you multiple usable reference views from a single generation.
Save the output as your master character reference. Every later prompt involving this character uses it.
- Also Read: “Best ID Card Makers Online For Free”
Step 4: Build a Vehicle or Product Reference Sheet
The same discipline applies to every other recurring element. For Slice Riot the vehicle is a red Shelby Mustang, appearing in roughly half the shots. Without a vehicle reference sheet, the car would shift in shape, proportions, and color from scene to scene.
Prompt: Create a complete cinematic car reference sheet based on the provided references.
Reference Target:
Image A: car exterior reference (use for exact model, shape, proportions, color)
Image B - C: car interior reference (use for dashboard, steering wheel, seats, materials)
Goal:
Generate a full, consistent car sheet that can be used as a master reference for future scenes.
Car Identity:
- strictly preserve the exact car model from Image A
- do not redesign or restyle the car
- maintain accurate proportions and details
- no modifications, no custom tuning
Exterior Views:
- front view
- rear view
- side profile
- 3/4 front angle
- 3/4 rear angle
Interior Views:
- driver seat perspective
- dashboard and steering wheel close-up
- center console
- back seat view
Style:
- realistic, cinematic
- neutral presentation (not dramatic scene)
- studio-like environment but not glossy commercial
Lighting:
- soft, neutral lighting
- evenly lit surfaces
- minimal shadows
- no stylized or neon lighting
Background:
- clean, neutral background (dark gray or black)
Camera:
- consistent lens across all views
- no distortion
- clean framing
Texture & Detail:
- realistic materials (metal, leather, plastic)
- accurate reflections (not exaggerated)
- no over-sharpening
- no stylization
Consistency:
- all views must match perfectly
- same car, same details, no variation
Important:
- no cinematic scene
- no motion
- no damage, no dirt
- no text
- no UI
How to adapt this for any product or object
The structure works for shoes, beverages, packaging, electronics, or anything physical that appears across multiple scenes. Keep neutral lighting and a clean background — reference sheets should look like studio shots, not finished ad frames. The "no cinematic scene, no motion, no damage, no dirt" lines are guardrails that prevent the model from turning your reference sheet into a dramatic shot before you are ready.
👉 Convert Your Text to Image Online with Zoviz’s AI Image Generator
Step 5: Compose Cinematic Hero Scenes Using Your References
Once character and product reference sheets exist, hero scenes become significantly easier. You pass the relevant references into the prompt and the model uses them as visual anchors — the practical foundation of a consistent AI video production workflow.
Here is the prompt for one of the key driving scenes in Slice Riot. It references three images: the driver face, the car interior, and a separate mood image dedicated to lighting. Lighting is part of identity too.
Prompt: Create a cinematic interior car shot from the back center area between the two front seats, looking forward through the windshield of a fast moving car.
Reference Target:
Image A: driver face, must match exactly, clean face, no injuries
Image B: car interior reference
Image C: night neon cinematic lighting reference
Framing:
- camera is positioned between the two front seats
- camera looks forward through the windshield
- driver is visible on the left side of the frame in semi-profile
- steering wheel and dashboard partially visible
- windshield is the main view
- road and traffic visible ahead through the windshield
Action:
- the car is moving fast
- the driver is tense and focused
- driver is actively steering
- traffic ahead feels close and dangerous
- our car is weaving through traffic
Motion Feel:
- strong sense of speed through the windshield
- cars ahead slightly smeared from motion
- fast light streaks outside
- tense, unstable driving energy
Lighting:
- low-key cinematic night lighting
- neon reflections across windshield and dashboard
- dashboard glow on the driver
- not commercial, not clean
Texture:
- subtle film grain
- slight softness
- natural shadow noise
- mild highlight bloom
Important:
- no pizza reveal yet
- do not show the back seat
- no face injuries
- no cuts, bruises, scars
- no text, no UI
What to take from this structure for your own AI image prompts for advertising
Reference targets always go at the top, with explicit notes on what each image controls. The framing block spells out camera position and what is or is not visible. The action block and motion feel block are separate — what is happening versus how it should read visually. Negative instructions close the prompt, blocking the most common unwanted insertions. This structure works across any category of cinematic hero scene.
Step 6: Edit Existing Frames Without Breaking Character Consistency
Sometimes a generated frame is mostly right but needs adjustment. The character is slightly off from the reference. The lighting style does not match the rest of the campaign. An earlier model produced a shot in a different aesthetic register that needs to be brought into alignment.
Edit prompts solve this without starting from scratch.
Prompt: Edit this existing frame and transform it into a cinematic, realistic night scene consistent with the rest of the project. (change driver)
Reference Target:
Image A: current frame to be edited
Image B: cinematic lighting & color mood reference (from the rest of the project)
Image D: driver character reference
Scene:
- nighttime urban environment, New York City
- ramp visible with car airborne
- car nose pointing upward
- emphasize vertical lift, height, and dynamic airborne motion
- driver visible inside, hands on steering wheel
Lighting & Mood:
- cinematic low-key night lighting
- subtle neon reflections and street light spill
- headlights illuminating path realistically
- reflections and ambient light consistent with reference mood
- subtle glow, practical cinematic shadows
Camera & Framing:
- maintain existing composition
- cinematic perspective emphasizing height and motion
- car fully readable, ramp and city visible for context
Texture & Realism:
- subtle film grain
- slight softness
- realistic shadows
- grounded reflection on car body and windshield
- avoid over-perfect AI textures
Style:
- cinematic
- gritty
- grounded realism
- practical stunt-film atmosphere
- match the tone of other project frames
- no UI
Important:
- preserve the layout and car position from the current frame
- integrate the driver naturally
- enhance realism and cinematic mood without changing composition
- no cartoonish physics
- no text
Why edit prompts matter in an AI video production workflow
"Maintain existing composition" and "preserve the layout" keep what already works. The lighting and mood reference image is the bridge between this frame and the broader campaign — without it, fixing one frame creates a new inconsistency. And instructions like "change driver" let you back-propagate a finalized character reference into frames that were generated before the character sheet was locked.
Edit prompts are what make full campaign consistency achievable even when scenes were produced at different times with different models.
Step 7: Generate the Social and UGC Content Layer
The cinematic hero scenes are one layer. The social and UGC content creation layer is where the brand meets its audience every day — and most AI campaigns underweight it significantly.
For Slice Riot the social ecosystem includes Instagram feed posts in multiple format variations, Stories and Reels concepts, carousel layouts, teaser and reveal sequences, and X-style posts in a deliberately rough aesthetic.
The UGC layer is built with different intent than the hero scenes. Where hero scenes lean polished, UGC leans imperfect. Grain and camera noise. Natural shadows and uneven lighting. Casual or off-center framing. The feel of someone's phone — not a campaign deliverable.
How to write AI commercial prompts for UGC content
Write UGC prompts in a different register than hero scene prompts. Less cinematic vocabulary, more casual description. Drop the "85mm lens" and "low-key cinematic lighting" language and replace it with "phone camera quality," "slightly overexposed," "off-center." The contrast between your hero scenes and your UGC layer is what makes the campaign feel like a real brand living in the real world.
Generate multiple variations of every key moment. Real audiences see several phone clips of the same situation — not one perfect frame. Mix demographics, locations, and contexts. Format X-style posts with realistic usernames, reply counts, and timestamps so they read as native content.
Step 8: Add Cinematic Motion With the Right Tools
Once still frames are locked, motion is the final layer of the AI video production workflow.
Kling handled the bulk of the motion-heavy sequences in Slice Riot — large-scale movement, reveal sequences, environmental interaction, and cinematic pacing. The priority here is believable motion and realistic camera behavior.
Seedance was reserved for hero reveals and cinematic transitions — the moments that need to land emotionally. Motion continuity, transition quality, and payoff timing are the priorities at this stage.
How to decide between motion models
Use the higher-volume model for routine motion: driving shots, environmental movement, location transitions. Reserve the higher-craft model for the three or four moments in the campaign that have to land — the hero reveal, the product moment, the closing frame. Pass the same reference images used for still generation into the motion model. Identity has to carry into motion or you lose everything built in earlier steps.
For complex sequences, build elements separately before compositing. Vehicles, environmental details, lighting interactions, and particle layers can all be generated independently and assembled together. Layered production produces results that feel grounded rather than synthetic.
👉 Create Thumbnails For Your YouTube Video That 3X Your CTR with Zoviz’s Youtube Thumbnail Maker
Common Mistakes That End AI Ad Campaigns Early
Skipping the brand identity step. If you start generating before you know what the brand looks and sounds like, every prompt invents identity on the fly. No two prompts invent it the same way.
Generating scenes before reference sheets. Every recurring element — character, product, vehicle, location — needs a reference sheet first. Without them, character consistency in AI image generation is not achievable. It is guesswork.
Overpolished UGC. UGC that is too clean reads as fabricated faster than anything else. Lean into grain, blur, imperfect framing, and casual composition. If it looks like a campaign asset, it has already failed as UGC.
Vague AI image prompts. "A cinematic shot of a man driving" produces a generic shot of a generic man. Every prompt in this article runs several hundred words. Specificity is what produces specific results.
No negative instructions. Models generate what seems plausible given the prompt, and they frequently insert elements you did not ask for. Tell them explicitly what not to include. Scars, beauty retouching, stylization, UI overlays, text — all of these need to be blocked in every hero prompt.
Mixing models without an editing pass. Different models have different aesthetic defaults. Using multiple models across one campaign without a consistency editing pass leaves visible seams. Build edit prompts into your workflow from the start, not as a last-minute fix.
Final Thought
The hardest part of producing a cinematic AI video ad campaign is not learning the tools. The tools change every few months. It is the discipline — committing to brand identity before touching a generation tool, building reference sheets before composing a single scene, and running consistency editing passes rather than shipping whatever came out.
The teams who produce coherent campaigns are the ones who treat this work like production, not experimentation. The Slice Riot example is concrete, but the workflow here scales to any brand, any category, and any budget. Start with one creative seed. Lock identity. Build the reference sheets. The rest of the pipeline becomes a series of clear, manageable decisions rather than one impossible creative problem.
👉 Watch the finished campaign: https://youtu.be/daEzpH3Sv6c
👉 Try the workflow in Zoviz Studio: https://zoviz.com
FAQ: How to make an AI commercial Video?
Do I need to use the exact models that Slice Riot used?
No. The production used specific tools available at that time. The workflow and prompt structure are model-agnostic and apply to any modern AI image and video generation pipeline.
How specific does my brief need to be at the start?
Less specific than most people assume. Slice Riot began with one sentence. Specificity comes in during the brand identity step, not the brief. Starting with a single clear concept is better than starting with a detailed document that has not been tested creatively.
How long do reference sheets stay useful?
For the lifetime of the campaign and beyond. A strong character or product reference sheet anchors follow-up campaigns, regional adaptations, reactive content, and social rollouts months later. Build them as long-term brand assets, not single-use generation inputs.
Can I use this workflow without a dedicated platform?
Yes. Every prompt in this article works in any pipeline that accepts text and reference images. A chat interface with MCP integration — such as Claude connected to Zoviz Studio — makes the workflow faster because orchestration happens in one conversation, but the techniques are fully tool-independent.
What is the difference between Zoviz Studio and the Zoviz MCP?
Zoviz Studio is the creative platform where campaigns are produced, organized, and exported. The Zoviz MCP is the connector that lets Claude call Zoviz capabilities directly from a chat interface. They work together: Claude handles creative direction, Zoviz Studio handles asset production.
How do we maintain consistency when multiple people are running prompts?
Centralize the reference sheets and the brand identity document. Every team member pulls from the same master sources before generating anything. Shared reference sheets are the most practical consistency tool available for teams, and the cheapest problem to solve before production starts.
