You have seen the videos. Someone's protein powder on a giant screen in Times Square. A bottle of face wash suspended in mid-air with water frozen around it. A juice bottle on a roadside billboard that looks like it cost twenty thousand dollars to shoot.
And underneath every single one, the same comment, hundreds of times: how do I make this?
The answer almost never comes. Instead you get "DM me". "Follow first and I'll send it." "Comment PROMPT and I'll reply." Thousands of people ask, a handful get an answer, and everyone else follows an account waiting for something that never arrives.
It works as a growth tactic, which is exactly why it is everywhere now. But here is the thing worth saying plainly: a prompt is a paragraph of text. It costs nothing to give away. Holding it back for a follow does not protect anything valuable — it just wastes the time of people who genuinely wanted to learn something.
So this article gives you the prompts. All of them, on the page, with an explanation of why each part does what it does.
What you need before you start
Less than you might think. You do not need Midjourney, a subscription, or any design skill.
- A photo of your product. A phone snap on a clean table is genuinely enough. The prompt supplies the lighting and the setting — but it cannot invent detail your photo never captured, so wipe the product first and shoot near a window.
- ChatGPT or Gemini. Both accept an uploaded image alongside a text prompt, which is what all of this depends on. Free tiers work.
That is the whole setup. Upload the photo, paste the prompt, wait.
The Times Square billboard prompt
This is the one everybody asks about. Here it is:
Photograph of a large illuminated LED billboard in Times Square at dusk displaying the attached product image. Camera at street level looking up at roughly 30 degrees, wide 24mm lens showing surrounding buildings and other lit signage. Wet pavement reflecting the screen light. Blurred crowd in the foreground with motion blur. Blue hour sky, screen bright enough to cast light on nearby surfaces. Realistic photographic perspective with the billboard slightly angled, not flat to the camera.
Why it works
Two clauses are doing almost all the work, and if you remove them the image falls apart.
"Looking up at roughly 30 degrees" gives the billboard real perspective. Without it, the model renders a flat rectangle pasted onto a street photo — which is exactly what a bad mockup looks like. Telling the model where the camera is standing is what makes the screen sit in three-dimensional space.
"Bright enough to cast light on nearby surfaces" turns the screen into a light source rather than a sticker. Real billboards illuminate the pavement, the crowd, the sides of buildings. That spill of light is what your eye uses to decide whether something is really there.
The dusk setting is not decoration either. A bright LED screen cannot compete with midday sun, so daytime billboard images always look washed out and grey. Blue hour gives the screen something to glow against.
The floating product prompt
The second most-asked one — the product suspended in mid-air with a splash frozen around it.
Dynamic commercial photograph of the attached product floating in mid-air, rotated slightly toward the camera. Frozen splash of water and fresh ingredients suspended around it, caught mid-motion. High-speed flash freezing every droplet sharply. Deep blue to teal gradient background with a soft glow behind the product. Shot on a 100mm macro lens, everything crisply in focus. Vibrant, high-contrast commercial colour grade.
Why it works
"High-speed flash" is the phrase that freezes the splash. Without it, models interpret "splash" as movement and render motion blur, which reads as a mistake rather than a technique. Real commercial photographers use flash durations measured in fractions of a millisecond for exactly this reason, and naming the technique gets you the result.
"Rotated slightly toward the camera" prevents the flat, straight-on rendering that makes floating products look pasted in. A tiny rotation gives the object volume.
One thing to change for yourself: match the suspended elements to the product. Coffee beans for coffee, citrus for drinks, oats and berries for cereal. Writing "ingredients" gets you generic ingredients, and the image stops being about your product.
The 3D billboard prompt
Photograph of a large roadside billboard against a clear blue sky, displaying the attached product with a bold headline. The product rendered so it appears to extend beyond the billboard frame in three dimensions, with a realistic drop shadow onto the billboard face. Steel support structure visible below. Camera positioned low and to the right, 35mm lens, natural midday sunlight from the left. Slight upward perspective. Clean, high-contrast commercial advertising style.
Why it works
"Realistic drop shadow onto the billboard face" is the entire 3D illusion, and it is the clause most people leave out. Anything that appears to break out of a frame has to cast a shadow back onto that frame. Without the shadow, your eye immediately reads the product as a flat overlay, no matter how good the rest of the image is.
Keep any headline to four or five words. Image models render short text reasonably well and long text badly — every extra word raises the chance of garbled letterforms.
The structure behind all of them
Once you have run a few of these, you start to notice they follow the same order:
subject → shot type → lighting → environment → lens → finish
This matters more than any individual prompt, because it lets you write your own. Image models handle that sequence far better than a pile of adjectives, since each clause narrows the one before it.
Compare these two:
- "A blue bottle, beautiful, professional, 8K, ultra detailed, masterpiece" — gives the model nothing to act on. Those words describe a feeling, not a photograph.
- "A blue bottle, three-quarter view, lit by a single softbox from the left, on a concrete surface, 85mm lens, muted colour grade" — describes a photograph that could actually be taken.
The second one wins every time, and it contains none of the words people think matter.
Two habits that improve every result
Name the light
Direction, quality and source. "Soft light from the upper left through a diffusion panel" is a real instruction. "Good lighting" is not. Lighting is the single biggest difference between a product photo and a snapshot, and it is the thing models respond to most reliably.
Say what you do not want
Image models decorate by default. They add gradients, glows, floating shapes, stray text. A clause like "no gradients, no decorative elements, no text" often improves an image more than another sentence of description would.
Why results vary between runs
Image models are not deterministic. The same prompt, run three times, gives three different images. This is not a fault in the prompt and it is not something you can fix by adding more words.
Run each prompt two or three times and keep the best. That is what everyone posting these images does, including the accounts charging you a follow for the privilege.
And if the results are consistently disappointing, look at your input photo before you rewrite the prompt. Clean, evenly lit reference images produce dramatically better output than clever wording ever will.
The full library
There are around fifty of these in our AI Prompt Library — product photography, billboard and packaging mockups, social media sets, food and drink, professional headshots. Every one includes the same breakdown: what it produces, why it works, what to change for your own product, and variations to try.
No email, no follow, no DM. Copy and go.
What to do with the image afterwards
Generated images tend to come out large and in the wrong aspect ratio for wherever you are posting them. Two of our free tools handle that: the image resizer for platform-specific sizes, and the background remover if you need the product isolated for a different composition.
A note on using these commercially
The prompts are free to use however you like, including for client work. What you may do with the generated images is a different question, and it depends on the terms of whichever AI tool you used rather than on us. Those terms change, so check the current ones for ChatGPT, Gemini or Midjourney before you sell the output or put it into a paid campaign.
One thing worth being careful about regardless: mockups showing your product in a real location like Times Square are illustrations of an idea, not evidence of a campaign. Presenting them as a real advertisement you ran would be misleading, and it is the kind of thing that damages trust far more than it impresses anyone.