Quick answer
The breaking-news AI video turns one or two photos into a fake live TV broadcast: you're at the top of the Empire State Building holding a hand-painted banner with your own message, shot like news-helicopter coverage with a reporter voiceover. You type the sign text (a proposal, a promo, a birthday, a flex), and AI renders the climbers, the banner, and the 'BREAKING NEWS' chyron. Most people prompt a native-audio model like Seedance 2.0, or skip the prompt with Starrd's Breaking News template — upload a photo, type your banner, and it renders the whole broadcast for 100 credits (about $5.99), no subscription.
What You're Trying to Make
One photo in, and out comes a fake live TV bombshell: you at the very top of the Empire State Building's spire, unfurling a hand-painted banner with your own message, caught by a circling news helicopter — a "BREAKING NEWS" chyron across the bottom, a breathless reporter narrating, rotor wash and wind in the background. You write the sign; the AI does the climb, the broadcast, and the audio.
It blew up in July 2026 after a real daredevil couple free-climbed the building and got engaged at the top — covered live as "TRESPASSERS CLIMB EMPIRE STATE BUILDING." Within hours, creators were newsjacking the exact broadcast with their own banners: proposals, promos, jokes. This guide covers the whole thing — the look, the recipe, and the one-tap way to make yours.
Fastest way — Breaking News on Starrd bakes the whole broadcast into one tap: upload a photo (or two), type your banner in Director's Notes, and it renders the climbers, the sign, the chyron, and the reporter audio for free — no prompt to write. Want the full method first? Read on. ↓
The Look That Makes It Work
Three things separate a real breaking-news clip from a flat one:
- A believable broadcast frame. A red "LIVE" bug, a generic news-channel logo, and a red-and-blue "BREAKING NEWS" lower-third chyron. Keep the network fictional — a made-up "Channel 9 News" reads as real without borrowing a trademark.
- A legible, hand-painted banner. The sign is the payoff, so it has to be readable: bold white hand-painted capitals on a dark banner. Short messages land best ("SHE SAID YES," "HI MOM," "GRAND OPENING SAT").
- News-helicopter camera language. Long-lens telephoto shake from a circling chopper, a cut to the field reporter in the open door, then a steady zoom that settles on the sign. The wobble-then-lock is what makes it feel like a live feed instead of a render.
The Broadcast Recipe
If you're prompting it yourself on a model with native audio (Seedance 2.0 or Veo 3), this is the shape that works — a three-shot chopper sequence. Feed it your photo(s).
A live breaking-news helicopter broadcast, three shots, one continuous feed. Main subject: you (from your uploaded photo), as the climber at the top of the Empire State Building's spire, holding a large hand-painted banner reading "SHE SAID YES".SHOT 1 (0–4s): a long-lens aerial from a circling news helicopter — heavy telephoto wobble — you small at the very top of the mast with the banner. Overlay a red "LIVE" bug and a "BREAKING NEWS" chyron: "TRESPASSERS CLIMB THE EMPIRE STATE BUILDING / MIDTOWN MANHATTAN".SHOT 2 (4–7s): hard cut inside the helicopter — door open, wind whipping, handheld — a field reporter (a different, generic person) in a headset with a mic talking urgently to camera, city rushing past behind.SHOT 3 (7–12s): cut back to the long lens and zoom ALL THE WAY IN, stabilizing into a rock-steady telephoto hold on the banner — the text fills the frame, sharp and legible; the banner ripples, one subject waves.Audio: helicopter rotor wash and wind throughout, faint city below, and a breathless field-reporter voiceover about the climbers. No music.
The banner text is the whole product, so give the model the cleanest shot at it: keep the message short and in CAPS, and let it render on a single clean frame first (a text-strong image model like GPT Image 2 nails this) before the video zooms in. Long paragraphs will garble — a punchy line won't.
Step 1 — Pick the Photo(s)
One clear, well-lit, front-facing selfie for a solo climb, or two for a couple (the proposal version). It's the identity anchor — the model keeps your face recognizable while placing you atop the building. Dark, blurry, or group photos are the main cause of a face that drifts.
Step 2 — Write Your Sign
Decide what the banner says. This is where the format earns its shares. The real climbers' earnest "power of love" banner got roasted online as cringe — so the trend rewards the funniest, most self-aware message, not the most heartfelt. Pick your lane:
- 🧠 Self-aware / main-character (what's actually going viral):
THIS IS FOR THE VIEWS·MY ALGORITHM CHANGED AGAIN·NO ONE READS THESE·PLEASE CLAP - 💍 Proposal / love:
SHE SAID YES·MARRY ME·HAPPY ANNIVERSARY - 🎉 Announcement / promo (the brand play):
GRAND OPENING SAT·TICKETS ON SALE·NOW HIRING·IT'S A BOY·WE LAUNCHED - 😂 Funny flex:
HI MOM·TEXT ME BACK·I TOLD YOU I'D DO IT
Step 3 — Generate (Prompt It, or Tap Once)
Path A — prompt a raw model. Paste the recipe into Seedance 2.0 or Veo 3 with your photo. Both generate synced audio in the same pass. Expect to iterate — the failure modes are a garbled banner (keep it short) and a reporter that inherits your face (specify a different, generic reporter).
Path B — tap the template. Breaking News is Path A with the prompt already written. Upload your photo(s), type your banner (and, if you want, the chyron and the reporter's tone) in Director's Notes, and generate. 100 credits (about $5.99), no subscription, credits never expire.
Breaking: Empire State
Climb the Empire State Building and unfurl a banner with YOUR message, caught live on a news-helicopter broadcast. Upload a photo, type your sign — proposal, promo, or a funny flex. No prompt to write. 100 credits (about $5.99).
Common Mistakes
- A wordy banner. Long messages garble. Keep it to a few big words in CAPS.
- A real network's logo. Borrowing a real news brand invites trademark trouble and can trip content filters. Use a made-up channel — it reads just as real.
- The reporter looks like you. Without a clear "different, generic person" instruction, the model can paint your face on the reporter. Call it out.
- No AI label. A fake breaking-news clip should always be disclosed as AI — don't let it read as a real event.
- A bad reference photo. Blurry or dark input = a face that drifts. Spend your effort here.
Related Reading
- Viral AI Video Trends (2026): The Monthly Roundup — every trend worth making this month, each with a one-tap template.
- Seedance 2.0: The Complete Guide — the model behind realistic reference-photo video and in-pass audio.
- Seedance 2.0 Prompt Guide — the framework behind prompts like the one above.
