Vesta Bomb: the AI product commercial for a hot sauce that doesn’t exist (every prompt included)

We invented a brand from nothing. Then we designed its bottle, extracted its logo, and shot a cinematic TV spot where Marilyn, an astronaut and Cleopatra all reach for the same sauce. This is every prompt on the board.

Most ads start with a real product. This AI product commercial started with a name.

Vesta Bomb is a hot sauce that exists nowhere except inside a Magnific Space. No bottle to photograph, no brand book, no client. We built all of it with prompts: the packaging, the logo, the typography, and a four-scene commercial that jumps from a 1962 backstage to the surface of the moon to ancient Egypt and lands on a coffee table in a modern living room.

If you came here to copy, jump straight to the prompts. If you want to see how the whole thing hangs together, keep reading. Everything below is verbatim from the board, typos included. Raw prompts work. We’re not going to pretend ours were polished.

The idea: one bottle, every era

The concept is a classic advertising archetype: the product that has always been there. Marilyn seasons her chicken wings with it before singing happy birthday to JFK. An astronaut carries it across the moon. Cleopatra pours it on a grape. Then the spot snaps back to the present, to a messy, cozy living room, because the joke only lands if the last shot is yours.

That structure did most of the creative work for us. What the prompts had to solve was harder: keeping one product identical across four completely different worlds, and moving between those worlds without a single hard cut feeling cheap.

Two techniques carried the whole project. Reference chaining and the hidden face.

Reference chaining: the product never drifts

Every scene prompt on the board points back to the same node: [vesta bomb edited]. That’s the hero bottle render, referenced by name in every image and video prompt that follows. The bottle in Marilyn’s hand, the bottle in the astronaut’s glove, the bottle on Cleopatra’s gold table: same reference, same product, zero drift.

This is the single most important habit for any AI product commercial. Generate your product once, lock it, and treat that node as canon. Describe the world around it as much as you want. The product itself is never re-described, only re-referenced.

The hidden face: instant recognition, consistent shots

Look at how the prompts handle Marilyn and Cleopatra. Neither face is ever fully shown.

“(her face is not visible, only a bit of her iconic hairstyle)”

“(we can only see some of her face, her eyeliner and hair, enough to recognise its her)”

A hand, a hairstyle, an eyeliner wing, a dress. Your brain fills in the rest, and the model never has to nail (or fumble) a famous face. The scenes read instantly, the character stays consistent shot to shot, and the spot keeps its mystery. Constraint as style.

One thing this technique does not do: make the transparency question go away. A realistic AI scene built around a real person is still exactly that, face or no face. Label it as AI-generated when you publish, and check the rules that apply in your market first.

Part 1: building the brand

Before any scene, the product. Four prompts took Vesta Bomb from name to full brand kit.

The hero bottle

Three sentences. Product, background, aesthetic. When you’re inventing packaging, over-describing is the trap: the shorter the brief, the more the model commits to one clean design instead of averaging five.

The label iteration

One targeted edit instead of a regenerate. The bottle survives, only the pepper illustration changes. This edited result becomes [vesta bomb edited], the reference node every later prompt points to.

Extracting the logo

Extracting the typography

This is the underrated move. Once the bottle exists, you don’t redesign the logo for the end card. You extract it from the packaging you already approved. Brand kit in two prompts, and the logo that closes the spot is pixel-faithful to the label on the bottle.

Part 2: scene one, Marilyn

The frame

The motion

What to steal here: the video prompt is a shot list, not a description. Open on the drop, rotate, zoom out to the hand, bottle down, microphone up, exit, hold the frame. Each beat is a sentence, in order, and the model follows the sequence because there is one.

And note the last three lines. No music. Only sound effects. Soft cheering audience in the far background. Sound design lives in the prompt. That distant crowd is what tells you where and when you are without showing a single face.

Part 3: scene two, the moon

The moon sequence is built from four stills before a single frame of video is generated.

Then the video prompt choreographs the camera through all of them:

This is storyboarding, the way it has always worked, just faster. Generate the keyframes as stills, get each one right on its own, then hand the video model a route between them: footprint, pan up, orbit the astronaut, land on the bottle. The stills are your storyboard and your quality control at the same time.

The last still is doing double duty. That earth reflection on the glass isn’t decoration. It’s the doorway to the next scene.

Part 4: scene three, Cleopatra

Four stills again, plus the earth-reflection frame carried over from the moon.

And the video prompt, which contains our favorite transition on the whole board:

Read the first sentence again. The camera dives into the earth’s reflection on the bottle, crosses space, and lands on an Egyptian temple. Two eras stitched together through the product itself. That’s the kind of transition an editor would build in post over an afternoon, described here in one line, because the reflection frame was planned two scenes earlier.

The other quiet trick: both the Marilyn scene and this one end the same way. Character exits, camera holds on the bottle. The product is the last thing standing in every world. That repetition is the spot’s grammar.

Part 5: scene four, back to now

The set change

“Same position of the composition. same frame.” The bottle doesn’t move between ancient Egypt and a living room. Only the world around it changes, which makes the time jump feel like a match cut instead of a scene change.

The whip transition

The end card

The final shot breaks the fourth wall on purpose: a drop of sauce hits the lens, the frame blurs to dark red, and the logo we extracted from the bottle back in part 1 resolves out of it. The end card isn’t a title slapped on in post. It’s generated from the same brand kit as everything else, so the loop closes exactly where it opened.

Seven things this board teaches about AI commercial prompts

1. One product node, referenced everywhere. [vesta bomb edited] appears in eleven prompts. That’s why the bottle never drifts.
2. Write video prompts as shot lists. One beat per sentence, in order. The model respects sequence when you give it one.
3. Storyboard with stills first. Get every keyframe right as an image, then choreograph the camera between them.
4. Hide the faces. A hairstyle and a hand say Marilyn better than a rendered face ever will, and hold up shot after shot.
5. Plan transitions inside the frames. The earth reflection was placed on the bottle one scene before it became a wormhole.
6. Prompt the sound. “No music. Only sound effects.” plus two or three specific sounds does more for atmosphere than any style keyword.
7. End every scene on the product. Characters leave. The camera stays. The repetition is the message.

Try it with your own impossible brand

Vesta Bomb took a name, a Space, and a stack of prompts you’ve now read in full. Your fake brand (or your real one) is the same distance away.

Open a Space, generate your hero product, lock it as a reference, and start writing shot lists. START MAKING

FAQ

Can you make an AI product commercial from scratch?

Yes. This spot was built end to end inside Magnific Spaces: packaging design, logo extraction, four cinematic scenes and an animated end card, all from prompts. Every prompt used is published verbatim in this post.

How do you keep a product consistent across AI video scenes?

Generate the product once, then reference that exact node in every subsequent prompt instead of re-describing it. In this project, every scene points to the same [vesta bomb edited] reference, which is why the bottle is identical from the moon to Cleopatra’s table.

How do you write a good image to video prompt?

Write it as a numbered shot list: one camera move or action per sentence, in chronological order, referencing the still frames you already generated. Close with sound direction. The video prompts in this post follow that structure exactly.

Do prompts need perfect spelling to work?

No. The prompts here are published exactly as they ran, typos and all. Models read intent. Structure and references matter far more than polish.

How do you show a famous historical figure in an AI ad?

This project never renders a full face. A hairstyle, a dress, an eyeliner wing and context do the recognition work, which keeps the character consistent between shots. It does not remove your obligations: a realistic AI depiction of a real person may need a visible AI-generated label where you publish (in the EU this is now required under the AI Act), and likeness and IP rights still need review before anything goes live.