PRODUCT
product
Sizzling Burger Macro — Food Commercial Shot
A steam-and-sizzle burger hero shot with a slow push-in, written to avoid the two things that break AI food video: hands and readable menus.
Prompt
Create an 8-second 16:9 food commercial video in one continuous shot. Extreme close-up of a cheeseburger on a slate board, steam rising, cheese slowly melting down the patty. Slow push in toward the burger. Warm tungsten key light, dark moody background, shallow depth of field. 35mm film, slight grain.
Why this prompt
Food is quietly one of the best-fit subjects for Omni: the hero object is static,
the “motion” is steam and a slow cheese melt — both continuous, low-degree-of-freedom
movements a diffusion model handles far more gracefully than a dancing human
(digit.in’s simple-vs-complex finding).
The camera does the storytelling instead: slow push in is the intensity-building
move from the documented push verb family
(DeepMind guide), and the
camera vocabulary post covers why pairing it with a
motionless subject is the recommended use.
Source tier: 🟡 Pattern-composed (medium confidence — built from documented camera vocabulary and verified failure modes, output not yet video-verified)
The tungsten-and-slate art direction isn’t arbitrary either — dark backgrounds hide
the soft edges where generated steam meets the scene, and one style reference
(35mm film, slight grain) plus one lighting cue is exactly the
one-anchor-per-prompt density the vocabulary post recommends.
How to tweak
- Dish swap: the skeleton generalizes —
ramen bowl, steam rising, chopsticks resting on the rim(resting, not held — see failure modes),pizza pull...actually no: cheese pulls need a hand or utensil. Stick to self-animating dishes: ramen, steak resting with juices, coffee being poured from off-frame. - Faster social cut: 5 seconds,
9:16, start the push closer. Food holds attention better than most subjects at short durations. - Breakfast light:
warm tungsten, dark moody→bright natural window light, airy white background— the daypart flip is two phrase swaps. - Add the sizzle sound in post. Omni generates video; the sizzle your viewers feel is an audio-layer decision, not a prompt line.
Common failure modes
- No hands, no bites. A hand lifting the burger or a bite shot is the most-wanted food beat and the most reliable articulation failure (failure modes). Frame the food alone; imply the eater.
- No menu boards or packaging text. Restaurant-branded wrappers and chalkboard menus hit the documented text-rendering failure (PixVerse) — letters smear into pseudo-writing that reads as instantly fake.
- Don’t stack food adjectives. “Juicy delicious mouthwatering succulent” is the
adjective-stack anti-pattern; density comes from specifics (
cheese slowly melting) not superlatives. - Real menu items need reference images. Like real houses and real products, your signature burger needs an uploaded photo and image-to-video (Atlas Cloud); text alone generates a generic — if handsome — burger.
Notes
- If this runs as advertising for a real menu item, the generated clip must not misrepresent the actual product — food-advertising accuracy rules predate AI and still apply; disclose generation where required.
- Output carries a SynthID watermark.
- Pairs naturally with the coffee pour prompt for a two-clip cafe sequence chained in Flow.
Sources
- Simple vs. complex motion: digit.in hands-on
- Push verb family: DeepMind Gemini Omni prompt guide
- Text rendering failure: PixVerse review
- Image-to-video identity: Atlas Cloud features overview
Related