Google Is Buying the AI Video Market One Nickel at a Time

Five cents. That's what Google charges for one second of 720p video from Veo 3.1 Lite, the new bottom tier that went live March 31 through the Gemini API. And today — April 7 — they followed up by slashing Veo 3.1 Fast by a third: 0.15/sec down to 0.10. Two price cuts in one week. Google isn't competing on video quality right now. It's competing on the electricity bill.

#Three tiers, one strategy

Google's video generation API now splits into clean lanes:

Lite handles text-to-video and image-to-video, landscape and portrait, at 4, 6, or 8-second clips. It matches Fast's generation speed at less than half the cost. The tradeoff is visible — softer textures, occasional micro-jitter in camera motion, subject coherence that wobbles past six seconds. Perfectly fine for social templates and rapid iteration. Not fine for brand hero content.

The tiered structure nudges developers toward a specific workflow: explore cheap, render expensive. Generate fifty variations at the Lite price point, pick the three that work, then re-run those through Fast or the full model for the final asset. At five cents per draft and forty cents per final, the economics make sense in a way they didn't when every generation cost the same.

#The per-second landscape

Here's where major video APIs sit right now:

• Veo 3.1 Lite: $0.05/sec (720p) — the new floor

• Kling 2.6: ~0.07/sec without audio, ~0.14 with — still the best bang-for-the-buck on raw output quality

• Veo 3.1 Fast: $0.10/sec (720p, effective today) — the mid-tier workhorse

• Runway Gen-4.5: 0.15–0.20/sec — the cinematography tool

• Veo 3.1 full: $0.40/sec — maximum quality, maximum cost

Kling has been the value play in generated video for months, but Google just undercut it by 30%. And Kuaishou's API still demands a pre-paid package — $4,200 minimum for direct enterprise access, three-month commitment. Google's barrier to entry? Swipe a card on AI Studio, generate your first clip in under a minute.

Runway occupies different territory entirely. Gen-4.5 produces the most filmic output of any model on the market — lighting, depth of field, camera language that actually looks directed rather than computed. But at three to four times the cost of Lite, Runway is increasingly the tool you save for final renders, not the one you burn through while exploring fifty variations of a product reveal.

Then there's the ghost in the room. Sora reportedly burned $15 million per day on compute before OpenAI pulled the plug last week. Users have until April 26 to export their projects, and then the service goes fully offline. The company framed the shutdown as pivoting toward "world simulation research for robotics." Read: they couldn't make the economics work. Google watched that implosion happen in real time and chose the opposite path — underprice the market, lock in developers, improve quality iteratively.

#Who this is actually for

Lite doesn't target the creative director producing a brand campaign. It targets the developer building an e-commerce platform who wants to auto-generate product showcase clips. The social scheduling tool that needs ten thousand preview thumbnails a month. The edtech startup that wants visual explanations for every lesson without hiring a motion graphics contractor.

At 0.05/sec, a 6-second clip costs thirty cents. Budget 3,000 a month and you generate ten thousand clips. Try that arithmetic on Runway — 9,000 to 12,000 for identical volume. The difference isn't incremental. It's the gap between "we can ship this feature in Q2" and "that's next year's roadmap."

Google's playbook here is familiar to anyone who's built on their cloud. Subsidize the bottom of the market until the ecosystem locks in. Cloud Storage, Maps API, BigQuery — same pattern every time. Make it cheap enough that developers stop price-shopping and start building integrations. Raise prices later, once switching costs are real. That strategy works because Google can absorb the margin compression longer than any competitor except maybe Tencent.

#What Lite gets wrong

I don't want to oversell the output. Compared to Kling's best generations or Runway's cinematic renders, Lite clips look like what they are: a budget tier.

Hair texture goes mushy. Fabric looks painted on rather than draped. Camera moves occasionally stutter where a better model would hold smooth. And text rendering in generated video — already unreliable across every model — breaks down faster here. If your use case involves legible on-screen text, no current video model is reliable, but Lite is especially bad.

These are real limitations. But they're also the kind that shrink with every release cycle. Veo 3.1 Fast looked roughly like current Lite eight months ago. The trajectory points one direction.

#Six months from now

Two cuts in seven days. That cadence accelerates — it doesn't slow. I'd expect a Lite quality bump by summer that matches today's Fast output at the same five-cent price point. If Kling responds with developer-friendly API access — no pre-paid packages, no quarterly commitments — the per-second cost of decent generated video drops below three cents by Q3.

The question stopped being whether generated video gets cheap enough to embed everywhere. It already did. The question now is whether quality converges fast enough that developers stop caring which tier they're running. Today, the gap between Lite and Gen-4.5 is impossible to ignore. But the gap between Lite and "good enough for the use case" already closed for a lot of applications.