Cindy Zhu.
← all free guides
Marketing & ads Higgsfield

The levitating product ad system (2 prompts)

Hey, it's Cindy ๐ŸŒฑ You know the ads where a drink hangs in mid air and ice, caramel and strawberries are frozen in orbit around it, all lit like a perfume commercial? That used to be a photographer, a stylist, a retoucher and a few thousand dollars. Now it's two prompts and a photo off your kitchen bench. Here's the exact system.
the principle

The one thing that makes these look expensive ๐Ÿง 

Most people prompt "floating product with fruit around it" and get a generic energy drink surrounded by random strawberries. It looks like a stock photo.

The versions that actually stop the scroll do two things:

  • ๐ŸงŠ The floating elements are the product's own ingredients. Iced coffee gets ice, an espresso splash, caramel and beans. Matcha gets powder clouds, whisked foam and leaves. Chocolate gets cocoa dust, broken squares and a melted ribbon. The orbit is a deconstruction of the product, not decoration.
  • ๐ŸŽจ The background is pulled from the product's own palette. Warm espresso brown behind the iced coffee. Deep forest green behind the matcha. Red-brown behind the chocolate. That's why it feels designed instead of assembled.

So before you touch a prompt, answer one question: what is this product literally made of, and what colour is it? Everything else follows from that.


clean it up

Step 1: turn any photo into a studio shot ๐Ÿ“ธ

Upload your product photo first (a phone photo on a bench is genuinely fine), then paste this into Higgsfield.

๐Ÿ“ธ the studio re-stage
Use the uploaded photo as the exact visual reference for the product. Do not
redesign, reinterpret or improve the product in any way. Keep the label
artwork, typography, logo, colours, material finish, lid or cap shape, and
body proportions identical to the reference. If any text on the label is
too small or blurry to read in the photo, keep it soft and unreadable
rather than inventing words.

Keep the product exactly the same. Change only the lighting and the
environment.

Re-stage it as a premium studio hero shot:

PRODUCT: [PRODUCT, e.g. a 250ml canned iced latte]
ANGLE: [straight-on eye level / 15 degrees above / slight three-quarter turn],
product upright and centred, full product in frame, generous negative space
above and below it.

BACKGROUND: seamless [PALETTE, e.g. deep espresso brown] gradient sweep.
Darkest in the corners, with a soft glow directly behind the product so its
silhouette separates cleanly from the background. No props, no set dressing,
no text.

SURFACE: [matte dark stone / wet reflective black acrylic / brushed concrete]
with a soft believable contact shadow and a short vertical reflection
directly beneath the product.

LIGHTING: large softbox key light from 45 degrees front-left and slightly
above. A narrow rim light on each edge of the product to trace its outline
against the background. A low backlight behind the product to lift the
background and glow through any translucent parts. Specular highlights are
soft and elongated, never blown out.

OPTICS: shot on an 85mm lens at f/4. The label is sharp edge to edge, the
background falls off gently. No lens distortion or bending on the packaging.

FINISH: clean commercial retouch, colour accurate to the reference photo,
no dust, no fingerprints, no scratches. Keep the original [glossy / matte /
frosted] finish.

Remove the original background, benchtop, clutter, hands and every other
object completely.

Output: one single product, photoreal, vertical 4:5, high resolution.
The important part is that first paragraph. Models love to "improve" packaging, which is how you end up with a beautiful can that says NUTRIENT BOSST. Telling it to keep blurry text blurry is the single highest-value line in the whole prompt.

save this table

The variable cheat sheet ๐ŸŽจ

Fill these in and both prompts adapt to any product. This is the table to actually save.

Product typeFloat thesePaletteNever use
Coffee, iced lattesIce cubes, espresso splash arcs, caramel ribbon, milk swirl, roasted beansEspresso brown to warm amberRandom citrus
Matcha, green teaMatcha powder cloud, whisked foam, ice, tea leavesDeep forest green to jadeBerries
Chocolate, cocoaBroken chocolate squares, cocoa dust, melted drizzle ribbon, hazelnutsDeep red-brown to warm blackAnything cool-toned
Juice, smoothiesFruit halves, water crown splash, mint leaves, seedsThe dominant fruit's own colourIce if it's not a cold drink
Skincare, cosmeticsCream swirl ribbons, water droplets, oil bubbles, one botanical sprig, gel shardsSoft sand, blush, or the formula's own toneFood, splashy juice
SupplementsCapsules, a loose powder plume, the source root or berry, a scoopEarthy amber or deep plumCartoon molecules
Tech, electronicsGeometric glass shards, brushed metal panels, light streaks, exploded internal components, fine dustCharcoal or graphite with one accent pulled from the device's screen or LEDNo fruit, no water, no food. The number one mistake
Fashion, accessoriesFabric ribbons, leather swatches, chain links, dust motes, motion streaks, gold flecksTonal to the material itselfIngredients of any kind
Candles, home fragranceWax shards, dried herbs, cinnamon sticks, smoke ribbons, drifting embersMuted terracotta, olive, warm smoke greyHard geometric shapes
Cleaning, householdWater sheets, soap bubbles, citrus peel, folded fabricClean blue-green or bright whiteDark moody backgrounds
Rule of thumb: pick 3 to 5 element types, not 8. Three named ingredients beats a list of ten every time, because the model spends its attention rendering three things properly instead of cramming.

the money shot

Step 2: levitate it and add the orbit ๐ŸŒ€

Take the clean studio image from step one, upload it, and paste this. This is the money shot.

๐ŸŒ€ the levitation prompt
Use the attached image as the exact reference for the product. The product
itself must not change: same label, same typography, same colours, same
proportions, same finish, same size in frame. Only the environment and the
elements around it change.

SCENE: the product levitates in mid air at the centre of the frame, tilted
about 8 degrees, with no stand, no hand, no wire, no visible support of any
kind. It hovers slightly above a dark reflective surface and casts a soft
elliptical shadow directly below it, plus a faint blurred reflection.

FLOATING ELEMENTS: [INGREDIENTS, e.g. clear ice cubes, arcs of espresso
mid-splash, a curling caramel ribbon, whole roasted coffee beans] suspended
in mid air, frozen in motion, arranged in a loose spiral orbit around the
product. Some elements sit behind the product, some cross in front of it near
the edges of frame. They vary in size with distance from camera.

SPACING RULE: no element touches, overlaps or crosses the product's label.
Keep a clear gap around the product roughly equal to one tenth of its height.
Elements never intersect each other. The centre of the composition stays
clean so the product reads instantly.

MOTION: the elements closest to camera carry a light directional motion blur
in the direction they are travelling. The elements near the product stay
sharp. A few fine droplets and small particles drift through the empty space
to fill it without clutter.

BACKGROUND: seamless [PALETTE] gradient, dark and moody, darkest at the
corners, with a soft glow behind the product. Nothing else in the scene.

LIGHTING: dramatic studio setup. Large soft key from front-left above, hard
rim lights on both edges of the product and on the floating elements so every
element gets its own bright outline. Low backlight glowing through the
translucent elements ([ice / liquid / powder / glass]). Deep controlled
shadows, high contrast, glossy commercial finish.

REALISM DETAILS: [condensation beads and a few running droplets on the
product surface] [wet specular highlights on the ice] [a fine dusting of
loose powder catching the rim light]. Light from the background bounces
subtly onto the product so it sits in the scene rather than looking pasted on.

OPTICS: 85mm lens at f/2.8, focus locked on the product label. Foreground
elements fall gently out of focus. Slight lens bloom on the brightest
highlights.

COMPOSITION: vertical 4:5, product occupying the middle third, clean negative
space at the top for a headline. Photoreal commercial advertising photography.

Short version, for when you just want to test an idea quickly:

โšก the short version
[PRODUCT] levitating in mid air, tilted slightly, with [INGREDIENTS] frozen
mid-motion in a loose orbit around it, nothing touching the product or its
label. Dark moody [PALETTE] gradient background, dramatic studio lighting
with rim lights outlining every element, glossy commercial finish, 85mm at
f/2.8, focus on the label, motion blur only on the elements closest to
camera. Photoreal, vertical 4:5.

what to block

The negative prompt ๐Ÿšซ

If your tool has a negative prompt field, paste this in. If it doesn't, add it at the end of the prompt after the word "Avoid:".

๐Ÿšซ paste into negative prompt
distorted label, warped text, misspelled text, invented brand name, altered
logo, changed packaging design, duplicate products, two of the same product,
extra bottles or cans in frame, hands, fingers, arms, human presence, visible
stand, wire, pedestal, string, table under the product, elements overlapping
or intersecting the product, elements clipping through each other, cluttered
composition, cartoon, illustration, 3D render look, plastic CGI surfaces,
oversaturated colours, blown highlights, harsh flash, flat frontal lighting,
busy background, background text, watermark, logo overlay, lens flare across
the label, out of focus product, cropped product, tilted horizon

the craft

The dials that actually change the output ๐ŸŽ›๏ธ

  • ๐Ÿ”ฆ Rim light is what sells levitation. A bright edge outline on the product and on every floating element is what makes them look lit by the same scene. If your image looks muddy, ask for "harder rim lights on both edges".
  • ๐ŸŒก๏ธ The background should be the product's colour, darkened. Not black. Black flattens everything. A deep brown, green or plum with a glow behind the product gives you depth for free.
  • ๐ŸŽฏ Depth of field is the anti-CGI tool. f/2.8 with focus on the label and foreground elements soft mimics a real lens. Everything sharp everywhere is the fastest tell that an image was generated.
  • ๐Ÿ’ง Condensation and wet highlights make a cold drink look genuinely cold. For skincare the equivalent is water beads or a glossy cream edge. For tech, a faint dust catch in the rim light.
  • ๐ŸŒ€ Motion blur belongs only on the nearest, fastest elements. Blur everything and it turns to soup. Blur nothing and it looks like a diagram.
  • ๐Ÿ“ The spacing rule is the difference between a hero shot and a collage. Explicitly saying nothing crosses the label fixes about half of all failed generations.
  • ๐Ÿชž Reflection plus shadow. A hovering object with no shadow looks pasted. Ask for both and it locks into the scene.

the reasoning

Why the two-step split works ๐Ÿ’ก

You're giving the model two jobs instead of one. Asking it to fix lighting, remove your kitchen, preserve a real label, invent floating ingredients and light them dramatically all at once means it does all five at about 60 percent quality.

Splitting it means step one only has to protect the product, and step two only has to build a world around a product that's already correct. By step two the model is copying a clean reference instead of reconstructing packaging from a blurry phone photo. Every extra thing you ask a model to fix at once is another chance for the label to drift.


troubleshooting

When it goes wrong ๐Ÿ”ง

  • ๐Ÿท๏ธ The label went weird or text is misspelled. Go back to step one and regenerate. Don't try to fix it in step two, the error compounds. Add "keep all text exactly as in the reference, and keep unreadable text unreadable", and shoot the source photo straight on with the label flat to camera.
  • ๐Ÿ’ฅ Elements are crashing into the product or each other. Cut down to three element types and add "elements are widely spaced with clear dark background visible between them, nothing overlaps the product". Crowding is almost always too many named ingredients.
  • ๐ŸงŠ It looks flat, plasticky or obviously CGI. In order of impact: add condensation or surface texture, drop to f/2.8 so the foreground goes soft, then add "subtle surface imperfections, real photographic grain, light bounce from the background onto the product".
  • ๐ŸŒช๏ธ Too cluttered, product getting lost. Ask for "only 5 to 7 floating elements total, large areas of empty dark background, product clearly dominant", then push them outward with "elements orbit near the edges of frame, the centre stays clear".
  • โœ‚๏ธ The product looks pasted on. It's missing shadow interaction. Add "casts a soft elliptical shadow on the surface below and a faint reflection, light from the background wraps around its edges". Check your rim light direction matches your key light.

make it move

Step 3: turn the two images into the video ๐ŸŽฌ

You now have two frames that share a product and differ only in environment, which is exactly what a start-frame and end-frame video model wants.

Load them into Seedance (inside Higgsfield, alongside Kling and Veo):

  • ๐ŸŽฌ Start frame: your clean studio shot from step one
  • ๐ŸŒ€ End frame: your levitating version from step two
  • โฑ๏ธ Duration: 5 seconds
๐ŸŽฌ the Seedance motion prompt
the product slowly rises and begins to hover as the ingredients drift into
orbit around it, elements floating outward in slow motion, camera arcs gently
around the product by less than 30 degrees, smooth continuous shot, no cuts
Keep the camera move under 30 degrees. This is the rule that matters. The moment the camera swings wide, the model has to invent the sides and back of your packaging, and that's where labels warp and logos smear. A small arc reads as expensive. A big one reads as broken.

Then speed ramp it in your editor. Start slow, accelerate through the middle, ease out. That ramp is doing more work than anything else in making it feel like a real commercial.


go further

3 bonus prompts to run next ๐ŸŽ

1. The splash hero (same idea, no levitation, more energy)

1 ยท the splash hero
[PRODUCT] standing on a wet reflective dark surface, a dramatic crown splash
of [LIQUID] erupting around its base and arcing upward, individual droplets
frozen mid-air catching a hard rim light, deep [PALETTE] gradient background,
85mm at f/4, high-speed flash photography look, product label perfectly sharp
and unchanged, photoreal.

2. The exploded view (great for tech, skincare and supplements)

2 ยท the exploded view
[PRODUCT] floating in the centre of frame with its [COMPONENTS or
INGREDIENTS] separated and suspended in a neat vertical stack above and below
it, each element evenly spaced with clear gaps, precise technical arrangement,
soft top light plus twin rim lights, deep [PALETTE] gradient background,
clean minimal composition with negative space, photoreal product photography.

3. The ingredient reveal (the one that makes people stop)

3 ยท the ingredient reveal
[PRODUCT] hovering with a single [KEY INGREDIENT] mid-fall directly beside
it, caught at the exact moment of impact with a small burst of [POWDER or
LIQUID] radiating outward, extreme detail on the texture, dark [PALETTE]
background, single hard key light from behind creating a bright halo, macro
lens, shallow depth of field, photoreal.

pick your level

How to get the most out of it ๐ŸŽ“

๐ŸŒฑ If you're just starting don't change the prompt. Change only the bracketed variables, run step one, then step two, and look at what you get. The prompt is doing the heavy lifting, and the temptation to shorten it is what breaks it.
โšก If you know your way around image models the highest-leverage edit is the element list. Run the same product with three different ingredient sets and three different background tones. One combination will be dramatically better, and it's usually the one where the elements share a colour family with the background instead of contrasting against it.

the honest bit

The honest bit โœ…

Small text on labels is still the weak point of every image model out there. If your product has fine print, an ingredients panel or a barcode, either shoot it at an angle where those aren't readable, crop them out of frame, or accept that you'll compose the final ad with your real product cut out and pasted over the generated background. That's a legitimate professional workflow, not a failure.
Also budget for rerolls. Three to ten generations per usable shot is normal, even for people doing this every day.

The links ๐Ÿ”—

๐ŸŽฌ Higgsfield: higgsfield.ai (Soul and Seedream for the images, Seedance for the video)
๐Ÿค– Claude: claude.com

Follow @cindiezhu for more AI tips every single day ๐ŸŒฑ