Sora 2 is here.

Hey gang,

I needed to reconnect with nature this weekend. It’s been a wild week in the AI world.

OpenAI just dropped Sora 2, and it's not the kind of update you scroll past. I've been testing it, and it's a little unsettling how good it is.

Here's what matters this week for your work…

It makes longer videos where the physics actually work and people look like real humans from start to finish.

Video is now fast, cheap, and easy to create. Think UGC ads, spokesperson ads, social content, b-roll + more.

We've had some great video models out for a while (like Veo 3, Kling 2.5, Wan 2.5, etc), but Sora 2 looks amazing because it finally cracks two big problems that even the best AI video generators couldn't handle:

👉 Realistic physics (how objects actually move, fall, or collide)

👉 Believable people (faces, hands, consistent characters and consistent motion)

Before this, objects passed through each other, people floated around, hands melted into blobs, and faces changed halfway through a clip. Sora 2 actually fixes this stuff. Objects fall and bounce the way they should. People move naturally with faces and hands that stay consistent. Scenes don't fall apart after a few seconds.

We've come a long way.

But the feature that's getting everyone excited is… Cameo.

You record a quick video of yourself (just turn your head a bit and say a few words) and Sora turns that into a digital version of you that can be dropped into any scene it generates!!

Once you create your Cameo, you control who gets to use it. You can make it public (entire internet), limit it to friends or followers, or just invite specific people. Your friends/ team can put you in their videos, you can put them in yours, or you can create videos with both of you in the same scene.

I've been testing this myself, and I need to be honest: it's amazing, but it's still pretty primitive. The voice isn't entirely accurate and the face can look uncanny.

That said, it's well and truly good enough to create social media clips, ads, product demos, hero videos, B-roll, etc. Anything short-form where you need yourself in multiple settings without doing a full production shoot.

Check out some examples here, here & here.

You can also do image-to-video. Upload a still image and Sora will animate it, though uploads of real people are blocked unless you use Cameo.

They also launched an iOS app alongside Sora 2 that hit #1 on the App Store in 48 hours. It's basically OpenAI's answer to TikTok, a scrollable feed of Sora-created videos with sound and dialogue. It's a full social app built around AI-generated content.

So what are the limits?:

Where it still breaks:

How do I access Sora 2?:

It’s free to use, but currently invite-only and limited to USA & Canada. If you're outside those regions or don't have an invite code yet, keep reading…I've got you covered below.

If you want to see me build a production-ready UGC ad with Sora, learn how to bypass the watermark and metadata, access it outside the US/Canada, and grab my invite code - I created a full guide and demo walkthrough video below.

OpenAI just added Instant Checkout to ChatGPT in the U.S.

You can now ask for product recommendations, browse Etsy listings, and buy directly in the chat. Shopify stores are coming next.

This is a glimpse of how AI could handle your entire sales funnel. I am extremely bullish on GEO at the moment!! I think it’s the natural future of online shopping and every business should be looking into it. i'm about to be so rich.

Anthropic released Claude Sonnet 4.5 this week. It can now run autonomously for 30+ hours on complex tasks without losing focus, up from 7 hours with Opus 4.

Not my forte, but this is the most powerful model for coding tasks yet. It scored 77.2% on SWE-bench Verified, beating GPT-5 and Opus 4.1. Error rates in coding dropped from 9% to basically 0%.

It's also now the best model at controlling your computer (opening apps, navigating browsers, filling spreadsheets, completing multi-step tasks without you touching anything).

If you're building automations, prototypes, or internal tools, this is a big deal.

Sonnet 4.5 is way better than Sonnet 4, but Opus 4.1 still has a slight edge for writing.

The prose feels more human and lived-in, like there's actual thinking behind the scenes rather than just pattern matching.

It also crushed a 4,000-word SEO article from one prompt, with scannable structure and proper formatting. First model that passed the "would I publish this?" test.

Pricing stays the same as Sonnet 4 ($3 per million input tokens, $15 per million output tokens). If you're just chatting in the Claude app, price won't matter much. But for high-volume work or automations, it adds up.

Verdict: If you're coding, running automations, or doing content work at scale, Sonnet 4.5 is the clear choice.

Available now via Claude app or the API.

Sora just made UGC creators obsolete.

I just created 17 new UGC ads in 1 hour for less than $10.

I'm walking through my complete Sora process: removing watermarks cleanly, scrubbing metadata, and creating UGC ads that are genuinely undetectable.

Plus, I'm dropping my invite code so you can follow along in real-time.

This is the difference between obvious AI slop and content that actually converts.

Watch the full walkthrough here (and grab my invite code while it's still active).

Alright, that’s a wrap for this week. Go make some ads ya filthy animal.

Catch you next week,

Remy :)