A toolkit of multimodal AI experiences powered by Google's Gemini API — image, video, audio, and text generation in one place.

My Google I/O 2026 keynote watch list — what indie builders should actually listen for

Google I/O 2026 starts Tuesday May 19 at 10:00 PT and runs through Wednesday May 20. The keynote opens with Sundar Pichai, hands off to Demis Hassabis and Josh Woodward for the Gemini/AI surface, and ends with the YouTube and Android sessions that the tech press will mostly cover.

For indie AI builders, the keynote is two hours where you should be watching for seven specific signals. Stock price, consumer demos, partnership announcements — most of it is irrelevant to whether you ship something different on May 21 versus a week earlier.

This post is my watch list, in order of how much it'll affect what I build next. Background context: I've been building GeminiOmni on the existing Gemini stack for two months and have skin in the game on every one of these.

1. The Gemini Omni per-second price

This is the single most important number to get out of the keynote. If Omni does ship as a unified video+audio model — which I think is likely — its per-second pricing decides whether indie video tools are still viable next month.

What I'm hoping for: somewhere around $0.20 per second of generated video, putting it between Veo 3.1 Fast ($0.15/sec) and Veo 3.1 Standard ($0.40/sec) on a price ladder that still leaves room for indie tier structures.

What I'm fearing: a "premium feature" framing with no published per-call price, or a $1+/sec rate that prices out everything except enterprise. Either of those means Omni is a consumer Gemini feature, not a developer surface, and indie builders should keep defaulting to Veo 3.1 Fast.

The leak from 9to5Google didn't include pricing language, which I read as neutral — Google doesn't usually disclose prices in app-string leaks. We'll find out during the developer track, probably in a single slide between minute 45 and 55 of the keynote.

2. The developer API rollout timeline

The second-biggest signal: how fast does the new model get into developer hands?

Listen for the exact phrase: "available today in the Gemini Developer API." That phrasing means the public API is open the moment the keynote ends, and indie builders can ship integrations the same week.

If you hear "rolling out to developers in the coming weeks" or "available initially through AI Studio with API access following," that's a 2-4 week gap before you can build on it. Plan your sprint accordingly.

Google's track record on this is mixed. Veo 3.1 Standard hit the public API on launch day. Imagen 3 came out a month after the keynote where it was announced. There's no rule.

3. Whether they say anything specific about indie/solo developer pricing

Google has been quietly favorable to indie developers since the original Gemini API rollout in late 2023 — the Free tier is genuinely generous, the model pricing has been below GPT-4 class for everything below Pro, and the API quality has been good enough to build on.

What I'm watching for at I/O: a "Gemini for Builders" type announcement that codifies this commitment. Something like a free monthly inference allowance, dedicated developer onboarding, or community-led promotion programs.

This is medium-importance for me — it doesn't change what I build, but it changes the user-acquisition pitch. If Google ships marketing support for tools built on its API, every "Built on Gemini" badge gets more valuable. If Google doesn't, the value stays where it is.

4. Any signal about Nano Banana 2's roadmap

Nano Banana was the surprise hit of 2025 — 200 million image edits in its first month, more new Gemini users than any other feature, and the model that put chat-based image editing on the map. Version 2 shipped earlier in 2026 with 2K native output and better text rendering.

I want to hear what comes next. Specifically:

Is there a Nano Banana 3 in the works? At what cadence is Google iterating?
Does it get folded into Gemini Omni or stay as a specialized model?
Are there enterprise features (style libraries, brand kits, batch processing) coming?

This affects /tools/nano-banana-edit directly. If Google ships an API change that the editor needs to adapt to, I'd rather know on May 20 than discover it through users hitting errors.

5. The Live API getting non-preview status

Gemini 3.1 Flash Live is the most underrated thing Google shipped this year and currently sits at "Preview" status with a "free of charge" pricing footnote. Both of those are temporary.

What I'm watching for:

Does Live get promoted to General Availability? If so, the SLA story improves and I can comfortably build production voice products on it.
Does the "free of charge" pricing survive GA? Almost certainly not, but the new price point matters. Anything under $0.005 per minute keeps voice indie-viable.
Does multi-modal Live become a thing — i.e., can I stream video in alongside audio? That would change what's buildable on top.

I expect Live to get promoted to GA and to add multi-modal input, with a low but non-zero per-minute price. If the price comes in north of $0.05/minute, the indie voice apps story gets harder.

6. Anything that changes the long-context economics

1M token context isn't free — I spent a whole post on the per-page cost math. Two things at I/O could change the economics:

Context caching pricing changes. Today, cached context costs about $1/M tokens/hour for Flash. If Google cuts this by half, "drop a PDF and keep chatting" becomes much cheaper to offer at scale. The Pro tier of /tools/pdf-chat gets healthier margins overnight.

A 2M or 10M context window. Gemini 1.5 Pro had a 2M tier briefly. If Google brings that back at Flash pricing, the use cases expand significantly — full library cataloging, multi-document research, enterprise corpus search. Indie tools can grow into bigger jobs.

Neither of these is rumored, but Google has surprised us before. Worth listening for "context" anywhere in the keynote dialogue.

7. The Imagen 4 successor

Imagen 4 has been around since late 2025 and is overdue for an iteration. If Google announces Imagen 5 (or rebrands it as part of the Omni family), the Nano Banana 2 vs Imagen 4 decision matrix shifts.

What I'd love to see:

Imagen 5 with native chat-based editing — if Imagen gets Nano Banana's conversational interface, the two-model story collapses to one.
A "Imagen 4 Lite" tier below $0.02/image — there's headroom and demand at the cheap end.
Imagen output with embedded provenance / SynthID watermarks on by default — this would actually be good for the industry, not bad.

If Imagen stays unchanged through I/O, my image-tool architecture stays unchanged. If Imagen gets the chat treatment, I rebuild /ai-image-generator and /tools/nano-banana-edit to share a single chat surface.

What I'm doing during the keynote

A few mechanical things, in case it helps anyone else:

I'll be watching the livestream with pnpm dev open in a second window, ready to update /tools/text-to-video and the homepage hero with the new model name within an hour of any announcement that matters.
I'm pre-drafting two blog posts: one for "Gemini Omni shipped as expected" and one for "Gemini Omni shipped but it's not what we thought." Whichever one is correct goes live on May 20.
I have a feature flag wired up that lets me swap the default video model on /tools/text-to-video from Veo 3.1 Fast to whatever's announced, without a code change. One config update, deploy in 90 seconds.
I'm not pre-committing to any specific feature build until I've seen the actual demos. The leak suggests video; the keynote might surprise us with something else entirely.

What I'm not watching for

Worth saying out loud:

YouTube AI features — interesting for content creators, irrelevant for indie tool builders unless they make video editors.
Android announcements — only matters if you ship a native Android app, which 95% of indie AI tools don't.
Workspace AI updates — Gmail/Docs integration is consumer, not developer surface.
Stock price impact — entirely uncorrelated with what's worth building.

If you're an indie builder and you find yourself watching the YouTube creator-economy segment, scroll forward. Your hour is better spent on the developer track.

I'll write up the actual results on May 21

This post commits me to a writeup on May 21 — what the keynote actually announced versus what I expected, what changed in my plans, and what I'd build differently. If the predictions in this post age badly, it'll be in the open.

If you're following along: tune into the Gemini segment of the keynote (usually 45-75 minutes in), watch for the seven signals above, and assume that whatever the press writes that afternoon is downstream of what the developer track actually said. The slides matter more than the screenshots.

Three more days.