😺 Google's new image model is BANANAS...

PLUS: Young workers lose 13% of jobs to AI + Smith controversy

Welcome, humans.

A Will Smith concert went mega-viral this week—but not for his music. Instead, Reddit detectives were convinced he'd used AI to fake entire crowds of adoring fans, complete with melted faces and six-fingered hands.

Turns out the truth is way weirder than fiction. As Andy Baio of Waxy reported, the crowds WERE 100% real (we've got the receipts), but Smith's team apparently fed the real concert photos into AI video generators to create this final 1 min clip. Then YouTube secretly applied its own AI “enhancement” on top, creating a double-dose of digital weirdness that looked straight out of a fever dream

So we've gone from Will Smith being the internet's favorite AI victim (RIP spaghetti video benchmark) to being accused of AI fakery himself. The guy just can't catch a break with AI! If only he waited a few days for “nano banana” to come out… more on that below!

Here’s what happened in AI today:

Google released Gemini 2.5 Flash Image with superior character consistency.
OpenAI says it improved ChatGPT mental health safeguards after lawsuit.
Claude launched “Claude for Chrome”, a new agentic browser research preview.
Stanford study found AI caused 13% job decline for young workers since 2022.

Advertise in The Neuron here

Google’s new AI image editor, Google 2.5 Flash (a.k.a Nano-Bananas) is here…

Remember when AI image generators couldn't keep a character's face straight between two pictures? Well, Google just dropped its much hyped new model (codename: nano-banana 🍌), now known as Gemini 2.5 Flash Image, that you can try for free on AI Studio.

The big deal? This model nails character consistency like NOTHING we've seen before. You can place the same character in a desert, then underwater, then at a disco—and they'll actually look like the same person. Wild.

Let us give you an example of why this matters: Here’s what happened when we tried to use ChatGPT Imagen to edit Grant’s face for our podcast YouTube thumbnails:

Even if you’ve never watched our podcast, you can probably tell that’s not at all what Grant looks like… they’re not even consistent among themselves! Though TBH, that first guy is pretty cute…

Here's what makes nano-banana special:

Character consistency that actually works: Google built a template app showing how you can keep characters looking identical across scenes.
Edit photos (or drawings) with just words: Their photo editing demo lets you remove people, blur backgrounds, or colorize photos using natural language…and this co-drawing demo lets you draw and ask AI to fix it.
Actual world knowledge: Unlike other image models, this one knows stuff—like how the co-drawing demo turns doodles into learning experiences.
Multi-image fusion: You can now merge multiple images; fx, you can drag and drop objects between images seamlessly with their home canvas template.

Oh, and at $0.039 per image (just round that up to $0.04 per image lol, or $40 per 1K images), it's surprisingly affordable for what you get.

The speed at which this is spreading is nuts. OpenRouter just added it as their first-ever image model (out of 480+ models!), fal.ai is bringing it to their developer community, and Adobe just integrated it into Firefly.

That's right—Adobe's now offering both their own Firefly models AND Google's Gemini, plus models from OpenAI, Black Forest Labs (who makes FLUX), Runway, Pika, Ideogram, LumaAI, and others. It's like the AI model wars just turned into an AI model peace treaty…

Why this matters: We're watching the shift from “my model is better than yours” to “here's a buffet of models—pick what works.” Adobe gets it. Instead of forcing creators into one ecosystem, they're becoming the Switzerland of AI creativity.

We’re not sure if this has something to do with the big labs giving up the ghost on the “one model to rule them all”, scale-pilled AGI at all costs mode of marketing AI, or just a succumbing to reality that people just want the best model that works for them, but it’s an interesting development!

Either way, it’s becoming increasingly apparent that the future of AI isn't about one giant model to rule them all—it's about having the right tool for each unique job.

FROM OUR PARTNERS

Your keyboard called—it’s taking a long weekend ⌨️✈️

We’ve been banging away on keyboards for 150 years. Until now, voice dictation hasn’t been reliable enough to change that.

Wispr Flow finally delivers the no-edit confidence we’ve all been waiting for:

4× quicker than typing. Dictate emails, docs, and DMs in real time and save precious hours every week.
AI auto-edits on the fly. Flow cleans filler words, fixes grammar, and formats perfectly as you speak.
Works inside every app with no setup. Fly through Slack notifications, give more context to ChatGPT, or brain dump into Notion.
Use it at your desk or on the go. Available on Mac, Windows, and iPhone.

“This is the best AI product I’ve used since ChatGPT.”
— Rahul Vohra, CEO, Superhuman

Give your hands a break ➜ start flowing for free today.

Prompt Tip of the Day.

Justine Moore (who is an EXCELLENT AI creator and VC) shared her tips for working with Nano Banana Gemini 2.5 Flash (never gonna get used to calling it that). Here’s what she said:

Try broad, conversational prompts: Instead of breaking things down, try “continue this story” and it will write narrative AND create matching images.
Combine multiple edits in one prompt: “Remove everything except the woman and mic. Make her a 3D animated character. Put her in a photorealistic office” worked perfectly in a single go.
Upload multiple images at once: The model can work across several images simultaneously for more complex projects.
Try product photography swaps: Easily insert your product into existing scenes without recreating the entire background.

And if you wanna pull a Will Smith (the video thing…not the other one lol), you can follow her workflow for how to turn still photos into a sweet product video.

Check out all of our prompt tips of the day from August here!

Treats to Try.

*Asterisk = from our partners. Advertise in The Neuron here.

Welcome to Anything

*Anything: Launch beautifully designed mobile apps and web apps from just a prompt, no coding or extra tools required. Start building with Anything.
Claude in Chrome is Anthropic's new browser agent that navigates websites, clicks buttons, fills forms, and handles tasks like scheduling meetings and drafting emails hands-free (demo)—if you’re a Max subscriber, join the waitlist here.
Wan Video has a new open AI model called Wan-S2V that lets you upload an image and audio file, and it creates a video where the person moves and speaks in sync with your audio (GitHub, HuggingFace)—try it here.
Library app Libby launched an AI feature called Inspire Me that provides book recommendations either from prompts or from your previous reads.
Letta is the coolest program we’ve heard of recently…it lets you build AI agents with persistent memory that remember your conversations and preferences, so you can create assistants that get better over time instead of starting fresh each chat.
Creem splits your SaaS revenue with partners automatically and handles all the tax headaches across 100+ countries—free trial, then paid plans.
Draw a fish evaluates your fish sketches in real-time, and when your drawing looks fish-like enough, it comes alive and swims in a shared digital tank with fish drawn by people from around the world—free to try.

Around the Horn.

OpenAI says it improved ChatGPT's mental health safeguards after it was sued by a family whose son committed suicide, and will train it to ground delusional users in reality, strengthen protections to prevent degradation over time (where users might initially get proper crisis help, but later receive harmful advice), and are planning direct connections to therapists, emergency services, trusted contacts, as well as adding parental controls for teens.
Anthropic reached a binding class action settlement with the authors suing it over copyright infringement, moving immediately to ask the Ninth Circuit Court of Appeals to pause their ongoing appeal while they finalize the deal.
Stanford researchers analyzing millions of payroll records discovered that AI has triggered a 13% employment decline among young workers (ages 22-25) in exposed occupations like software development and customer service since late 2022, because AI can replace the "codified knowledge" from formal education that new graduates offer while struggling to substitute the tacit, experience-based skills that older workers possess.
The new AstaBench leaderboard tests AI agents on real scientific research tasks like literature review, data analysis, and code execution, revealing that even the best agents (Asta v0 at 53%) still struggle significantly with complex scientific workflows, especially data analysis where no agent scored above 34%.

FROM OUR PARTNERS

Context engineering for AI code reviews

Get context-aware code reviews on PRs or in your IDE with CodeRabbit. Using a multi-layered approach to context preparation, CodeRabbit suggests 1-click fixes and explains the reasoning behind suggestions, cutting code review time and bugs in half.

Learn more or try CodeRabbit today!