Skip to main content
Back

Google Gemini Omni Flash: edit any video with text prompts

Google drops Gemini Omni video model, GitHub Spec Kit hits 95K stars, grep beats vector search, ByteDance ships 3B multimodal model

May 20, 2026 6 min read

Hey,

Google I/O just dropped, and the theme is clear: AI is no longer a single assistant. It's a team.

Gemini Omni collapses the old relay race of separate video, image, and audio models into one. Antigravity 2.0 does the same for dev workflows, running parallel subagents like a full engineering crew. One model, one platform, many agents working at once.

We're watching the "single prompt, single output" era quietly end.
Together with · partner_image
Author photo
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
Summary

Read time: 6 min 41 sec

Top News

▸ Google DeepMind drops Gemini Omni, a model that edits and generates video from anything

Slack

▸ Building for the AI Era: Scale Your BI

Top News

▸ Google ships Antigravity 2.0 desktop app with multi-agent teams and native voice

Agent Field AI

▸ A new primitive for composing AI agents.

Top Repo

▸ GitHub's open-source Spec Kit hits 95K stars by making AI plan before coding

Signals

1. DeepSeek builds a custom VPN obfuscation plugin in 40 minutes flat
2. Presented by Wispr Flow: Teams at OpenAI and Vercel speak their AI prompts instead of typing them. Wispr Flow. 4x faster. 89% zero edits.
3. New book covers every RAG technique with comparisons, free on Kindle Unlimited
4. New 10B 3D reconstruction model cuts GPU memory 70%, beats all benchmarks by wide margin
5. Study finds grep beats vector search in agentic retrieval across multiple agent setups
6. ByteDance drops open-source 3B model that reads and generates text, images, and video
Top News
Google DeepMind drops Gemini Omni, a model that edits and generates video from anything
8,405 Likes
Google DeepMind drops Gemini Omni, a model that ed

Google just shipped Gemini Omni Flash, a model that takes any input (text, image, audio, video) and turns it into video. Not just pretty clips either. It actually understands physics, culture, and context.

Think of it like this: previous tools generated video from scratch. Omni lets you hand it a video you already shot and reshape it through plain conversation. Each edit builds on the last.

Here is what you can actually do with it:

  • Edit existing video with text: describe a change and Omni applies it, no timeline scrubbing needed
  • Consistent characters: define a person once, drop them into any scene, lighting and actions stay coherent
  • Mix inputs freely: combine a photo, a voice clip, and a text prompt into one output
  • Clips run up to 10 seconds today, with longer durations coming

Access is live now in the Gemini app, Google Flow, and YouTube Shorts. API access rolls out in the coming weeks. An Omni Pro tier is also in the pipeline for heavier workloads.

Read more
forward →
Presented by Slack
Building for the AI Era: Scale Your BI

You've deployed AI everywhere, but each agent generates isolated intelligence, deepening silos instead of breaking them. Your teams spend hours context-switching just to piece together the full picture.

Built from hundreds of Slack customer conversations, this playbook reveals the structural gap keeping intelligence trapped and shows you how to fix it.

Download four hands-on worksheets to diagnose gaps, trace decisions, and design your first pilot.

Read the playbook
partner with us →
Top News
Google ships Antigravity 2.0 desktop app with multi-agent teams and native voice
8,363 Likes
Google ships Antigravity 2.0 desktop app with mult

Google just shipped Antigravity 2.0, and it is a bigger deal than it sounds. This is not a plugin update. It is a completely separate desktop app, rebuilt from scratch, with one goal: let multiple AI agents work on your project at the same time.

Think of it like hiring a team instead of one assistant. You give a high-level goal, and Antigravity splits the work across parallel subagents, each handling a focused piece, without blocking each other.

Here is what you actually get:

  • Run multiple agents in parallel on the same project
  • Schedule background tasks that trigger agents automatically, no manual prompting
  • Export full projects from AI Studio to local Antigravity in one click
  • Native voice commands, Google Workspace API access, Android and Firebase integrations
  • A new CLI for terminal lovers, plus an SDK to build and deploy your own custom agents

Powering it all is Gemini 3.5 Flash, which Google says runs 4x faster than competing models. Download it now and try it.

Read more
forward →
Presented by Agent Field AI
A new primitive for composing AI agents.

The primitive that turns Claude Code, Codex, and Gemini from terminal tools into composable services.

Why your repo is the largest prompt you'll ever write. Why your boundary drifts mid-run.

How to engineer harnesses that hold in production. 100+ agent recipes on GitHub, Python/TS/Go, Apache 2.0.

Read the blog
partner with us →
Top Repo
GitHub's open-source Spec Kit hits 95K stars by making AI plan before coding
2,719 Likes
GitHub's open-source Spec Kit hits 95K stars by ma

You know that feeling when you ask an AI to build something and it just starts writing code immediately, skips half your requirements, and breaks everything? That's vibe coding , throwing prompts at an AI and hoping for the best.

GitHub just shipped Spec Kit to fix that. The idea: force the AI to fully understand what you want before it touches a single line of code. It's called Spec-Driven Development.

Here's the workflow you get:

  • /speckit.constitution , set your project rules and standards
  • /speckit.specify , describe what you want to build
  • /speckit.clarify , let the AI ask questions before starting
  • /speckit.plan , pick your tech stack and architecture
  • /speckit.tasks , break everything into ordered steps
  • /speckit.implement , actually build it

It works with 30+ AI agents including Claude Code, Cursor, Copilot, and Gemini CLI. It's fully open source, MIT licensed, and already sitting at 103k stars. Less debugging, more shipping.

Read more
forward →
Signals
1
DeepSeek builds a custom VPN obfuscation plugin in 40 minutes flat
2,473 Likes
2
Teams at OpenAI and Vercel speak their AI prompts instead of typing them. Wispr Flow. 4x faster. 89% zero edits.
Presented by Wispr Flow
3
New book covers every RAG technique with comparisons, free on Kindle Unlimited
27,382 Stars
4
New 10B 3D reconstruction model cuts GPU memory 70%, beats all benchmarks by wide margin
599 Likes
5
Study finds grep beats vector search in agentic retrieval across multiple agent setups
354 Likes
6
ByteDance drops open-source 3B model that reads and generates text, images, and video
513 Likes
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.

Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
Looking to promote your company, product, service, or event to 250,000+ AI developers?
Work with us