In partnership with

Welcome, humans.

It's Friday, so you know what that means: time to bust out the video memes.

For the past few weeks, social media dominated by memes from ByteDance's SeeDance 2.0, like these guys taking their dearly departed out for an F1 race:

Are we sure this is AI though??

Since this is The Neuron, we'd be remiss if we didn't share the best AI video cat meme of the week:

AI, why are you so scary sometimes?

It's not just your car starring in a horror movie; cats can get pretty horrified themselves!

I can promise you are not expecting this reveal…

Well, enjoy it while it lasts, SeeDance… because today's main story is all about Nano Banana 2 peeling out onto these streets faster than those coffin dancers. And idk if y'all clicked that, but those dudes were CRUISIN'!

Here's what happened in AI today:

Google launched Nano Banana 2, combining its best image quality with its fastest speeds across the Gemini app, Search, and Ads.
Claude hit #4 on the US App Store, an all-time high, with free users up 60%+ since January and daily signups tripling since November.
Anthropic rejected the Pentagon's "final offer" on AI safeguards, with a Friday deadline looming.
Apple agreed to pay Samsung double for iPhone memory chips because AI server demand is eating all the RAM.

Advertise in The Neuron here!

Google's Viral AI Image Generator Just Got a Major Upgrade (And It's Free Everywhere)

DEEP DIVE: Everything to know about Gemini's Nano Banana 2

If you've used AI image generation, you know the tradeoff: fast models look bad, good-looking models take forever. Google says you don't have to choose anymore.

Today, they launched Nano Banana 2 (officially Gemini 3.1 Flash Image), combining the quality of their Pro model with Flash speed. The original went viral last August. Pro followed in November. Now both live in one model.

Here's what's new:

Real-time world knowledge: It pulls from Google Search to accurately render real subjects, current events, and factual infographics, rather than guessing from training data alone.
Better text in images: Text accuracy across multiple languages is way better, making it actually useful for mockups, social posts, and greeting cards.
Subject consistency: Keeps up to 5 characters and 14 objects looking the same across a multi-image workflow (storyboards, ad campaigns, etc.).
Tighter instruction following, 4K resolution, and a visual fidelity upgrade (richer textures, sharper details, vibrant lighting).

Where it's rolling out: Nano Banana 2 is now the default in the Gemini app, Google Search across 141 countries, Google Ads, and the Gemini API. The easiest place to test it (for free) though? AI Studio; try it here.

Also, in case you missed our previous Banana coverage, every image gets a SynthID watermark plus C2PA Content Credentials so you (and everyone else) can verify it's AI-made. Pro/Ultra subscribers keep access to Nano Banana Pro for high-fidelity tasks.

Personal note: we use Nano Banana to generate The Neuron's header images, and consistency has been our biggest headache. Our cat mascot doesn't always look the same between editions, even with reference images. The subject consistency upgrade is the feature I'm most excited to test. Google also published a prompting guide worth bookmarking if you want better results; more on that before. Our first test, today's header, turned out great, and in one shot!

The question is: can it be used to create banger shorts like this one about our dystopian future? With Flow, it can!

And if you combine Google's AI video tools with Adobe's new fast AI editor, you'll be cranking out videos faster than Elon's hair just grayed.

FROM OUR PARTNERS

Free, private email that puts your privacy first

A private inbox doesn't have to come with a price tag—or a catch. Proton Mail's free plan gives you the privacy and security you expect, without selling your data or showing you ads.

Built by scientists and privacy advocates, Proton Mail uses end-to-end encryption to keep your conversations secure. No scanning. No targeting. No creepy promotions.

With Proton, you're not the product — you're in control.

Start for free. Upgrade anytime. Stay private always.

Get free private email

Prompt Tip of the Day

Google published a prompting guide for Nano Banana that's worth bookmarking. The key insight: simple prompts still work, but specific prompts unlock way better results.

The framework: think like a photographer, not a prompt engineer.

Every good image prompt has up to six elements:

Subject: Who or what's in the shot? Be specific. ("a stoic robot barista with glowing blue optics," not "a robot")
Composition: How's it framed? (extreme close-up, wide shot, low angle, portrait)
Action: What's happening? (brewing coffee, mid-stride running, casting a spell)
Location: Where? (a futuristic cafe on Mars, a sun-drenched meadow at golden hour)
Style: What's the aesthetic? (3D animation, film noir, watercolor, photorealistic, 1990s product photography)
Editing instructions (for modifying existing images): Be direct. ("change the tie to green," "remove the car in the background")

For pro-level results, add camera and lighting details like a cinematographer would: "A low-angle shot with shallow depth of field (f/1.8), golden hour backlighting creating long shadows, cinematic color grading with muted teal tones."

For text inside images, explicitly state what text should appear and how: "The headline 'URBAN EXPLORER' rendered in bold, white, sans-serif font at the top."

And for character consistency (the feature I'm personally most pumped about), upload reference images and clearly define each one's role: "Use Image A for the character's pose, Image B for the art style, and Image C for the background environment."

Best part of all this? You can upload this guide, and these instructions, into the AI and ask it to help you make the perfect Gem / Skill (packages of text and sometimes code telling the AI how to do something) to create prompts for you.

Treats to Try

Claude Cowork now handles scheduled and recurring tasks automatically, turning it from chatbot into operational agent. Set it to run reports, file organization, or updates on a schedule.
Nano Banana 2 in Flow lets you generate and edit images using Google's newest model for zero credits, making it the easiest free entry point for AI image generation.
Vinext is Cloudflare's open-source Next.js alternative, rebuilt from scratch on Vite in one week by a single engineer plus AI. Deploys to Cloudflare Workers.
Koah lets you place ads directly inside AI chat conversations across 50+ apps, reaching 500M+ daily GenAI conversations (raised $5M).
Tasklet just launched Instant Apps — tell your agent "build me a dashboard combining my Salesforce deals and Linear issues" and it creates a live, interactive app in seconds, connected to your real data with two-way sync across thousands of tools.

Around the Horn

I REPEAT, AI, WHY ARE YOU SO SCARY SOMETIMES?! I will say, SeeDance does really well with giant monster fare

Anthropic rejected the Pentagon's "final offer" on AI safeguards, with a Friday deadline looming; Secretary Pete Hegseth gave Anthropic until Friday at 5:01 PM to let the Pentagon use Claude for "all lawful purposes" (no restrictions), but Anthropic CEO Dario Amodei said no, arguing the company won't allow its AI to be used for mass surveillance of Americans or fully autonomous weapons.
Jack Dorsey halved Block's staff via 4,000+ AI-fueled layoffs, predicting most companies will follow within a year. Stock rose 24%. More of this type of activity by executives will invite plenty of political backlash, FYI.
Perplexity now powers 2 of 3 assistants on every Samsung Galaxy S26, the first non-Google company with OS-level access.
Apple expanded its Houston operations to manufacture AI servers and Mac minis in the U.S., with a new AI and automation training push.
1. Related: Apple agreed to pay Samsung double for the LPDDR5X memory chips it needs for iPhone 17 production after Samsung opened with a 100% markup as a "negotiating tactic," and Apple immediately accepted. The reason: chipmakers are redirecting capacity toward AI server memory, leaving mobile DRAM in extreme shortage.
NVIDIA's CEO Jensen Huang said markets "got it wrong" about AI killing SaaS, arguing agents will use software rather than replace it.
Amazon's AGI lab leader David Luan left the company to "cook up something new," adding to the talent churn war between frontier AI labs.
Claude hit #4 on the US App Store (an all-time high), with free users up 60%+ since January and paid subscribers more than doubling since October. The company credits Claude Code, Cowork, and word-of-mouth driven by Claude's character and tone.
A new study tracked 2,430 responses from Claude Code and found it overwhelmingly builds custom solutions rather than recommending third-party tools. When it does pick a tool, it picks hard: GitHub Actions 94%, Stripe 91%, shadcn/ui 90%. Notable: zero picks for Express, Redux, or traditional cloud providers like AWS/GCP/Azure for deployment. (Full report)