There’s a new trick in the kit that makes it easier for developers — and anyone they build for — to drop AI-made clips straight into the vertical scroll. Google announced that its Veo 3 video model (and its cheaper sibling, Veo 3 Fast) now supports 9:16 vertical output and higher resolution, plus big price cuts that make mass generation a lot less painful for wallets. If you make — or consume — short videos, this changes the playing field.

What changed

• Vertical 9:16 output. Developers can now set aspectRatio to 9:16 in the Gemini API to produce portrait videos built for phones and social feeds (TikTok, Reels, Shorts). That means you no longer have to generate widescreen and awkwardly crop or reframe it yourself.
• 1080p is here. Veo 3 supports 1080p resolution now — a solid step up from the previous 720p cap. (A few outlets note a current restriction: 1080p generation is functioning for 16:9 outputs today, not yet for every orientation.)
• Much cheaper to run. Google reduced prices across the board: Veo 3 is now $0.40 per second (down from $0.75) and Veo 3 Fast is $0.15 per second (down from $0.40). Google says both models are “stable and ready for scaled production” in the Gemini API.

Those are the bullet points; underneath them are some practical and ethical consequences that matter.

A quick developer checklist (what to change in your app)

If you’re integrating Veo 3:

• Use the Gemini API’s aspectRatio parameter and set it to 9:16 portrait clips.
• If you need sharper output, set resolution to 1080p — but test if your use case demands 16:9 (the 1080p limit has been reported as tied to that ratio for now).
• Recalculate costs: example math below. (Arithmetic shown step-by-step so you can trust the numbers.)

Example cost math (audio + video, new prices):

• An 8-second clip with Veo 3: 8 × $0.40 = $3.20.
  • Step: 8 × 0.40 = 8 × (4 ÷ 10) = 32 ÷ 10 = 3.20.
• The same clip with Veo 3 Fast: 8 × $0.15 = $1.20.
  • Step: 8 × 0.15 = 8 × (15 ÷ 100) = 120 ÷ 100 = 1.20.

So an eight-second vertical Short could now cost you a few dollars to generate at Veo 3 quality, or roughly a dollar-plus with the Fast model — a gulf that matters if you’re batch-generating hundreds or thousands of clips.

Why Google probably pushed this now

Short-form, vertical video dominates attention on phones. Google has been explicit that Veo will be baked into its ecosystem — most notably YouTube Shorts — which will make Veo-generated content native to one of the largest short-video pipelines on Earth. That integration was foreshadowed earlier this year and now these developer-facing additions make that rollout more practical.

Lower prices do another thing: they let experimentation scale. At $0.40/sec, small studios and indie developers can prototype more iterations; at $0.15/sec, large-volume industrial use (automated content, dynamic ads, localized clips) becomes cost-plausible.

The creative upside (and the inevitable “AI soup”)

Veo 3’s vertical output + higher res means creators and apps can:

• Generate social-native footage directly (no post-crop, fewer artifacts).
• Make regionalized or personalized Shorts at scale (localized product videos, weather visuals, dynamically generated UGC).

But — and it’s a big but — a more accessible generation will also turbocharge the quantity (and noise) of AI videos. Expect a wave of novelty content, cheap ads, and experimental storytelling — and more filler pumped into already crowded feeds. The phrase “AI slop” used by critics isn’t hyperbole: easier tools mean more low-effort output will hit eyeballs.

Safety, detection and the messy reality of misuse

Google has built mitigation tools — including its SynthID watermarking technology and a visible Veo watermark in many clips — to help label AI-generated media. But the safeguards aren’t a perfect silver bullet. Researchers, journalists and watchdogs have already documented Veo-generated clips that violated policies or spread harmful content, and visible watermarks can be small, missed, or removed after the fact. That mismatch between capability and moderation capacity is the central tension here: the tech is outpacing the systems that police it.

What that looks like on the ground:

• Watermarks & provenance: Google embeds SynthID (an invisible provenance marker) and, in many cases, a visible tag; that helps detection and research, but won’t stop a determined abuser.
• Real misuse has happened. Reports have surfaced of racist or otherwise policy-breaking AI videos generated with Veo that went viral on platforms like TikTok, forcing removals and account bans. Those incidents make clear that detection + human moderation still need to scale.

What publishers, platforms and policymakers should watch

• Authenticity standards. Platforms will need clearer display rules: when does a visible watermark suffice? Do we require labels at the player level? Who enforces it? Recent debates suggest industry standards (like C2PA / SynthID-style provenance) will be part of the answer, but not the whole answer.
• Moderation scale. Automated filters will catch some abuse, but human reviewers still matter — and those teams must be ready for new classes of content that are realistic, dynamic, and hard to trace.
• Economic incentives. Lower prices encourage novelty and volume. That’s great for creativity, but it also means bad actors can test, iterate, and amplify at low cost.

Google has made Veo 3 better at the things people care about for mobile: portrait framing and sharper image quality — and it’s much cheaper to run. That means more AI-generated vertical clips are coming to our feeds, faster and cheaper than before. It also means that the arguments about labeling, moderation, and the ethics of generated media are not academic anymore — they’re immediate, practical, and already playing out on the platforms we use every day.