This was a FAST Week (and Unexpected!)

Hey gang, welcome back to Theoretically News!

Here's a sentence I did not expect to write this week: Midjourney, the image company, just built a full-body medical scanner? That was unexpected.
We've also got LTX and Luma both leveling up, and a genuinely delightful little time passer to check out!

Let's dive in!

File under “Things NO ONE Expected”
After teasing "first hardware" all week, Midjourney pulled back the curtain — and it's not THE ORB, or The Egg.

Instead, it's the Midjourney Scanner: a full-body ultrasonic CT machine built from a ring of hundreds of thousands of ultrasound transducers, no radiation. You step onto a platform, get lowered into a shallow pool, and a ring of sensors images your insides on the way down. Founder David Holz is calling it the first genuinely new whole-body imaging method in 50 years.

The numbers come with an asterisk. Midjourney claims 10× cheaper than an MRI and a 60-second scan — but the current prototype takes around 20 minutes, the team is nine people, only a dozen-ish humans have been scanned, and there's no FDA clearance. Holz admitted it barely uses AI yet.

It's built on an ultrasound-on-chip license from Butterfly Network, signed quietly back in November — an eight-figure deal. The roadmap's just as bold: a "Midjourney Spa" in San Francisco by the end of 2027, scaling toward 50,000 scanners and a billion scans a month

My Take: Overall, I think it's cool. There's been pushback from the established medical community, and a few raised eyebrows reading this as either a cash grab or a pivot to something more profitable. I'm not remotely qualified to weigh in on the medical side — but as a move, this is genuinely interesting, and honestly one I wish we saw more of.

Midjourney is in a unique spot for this kind of side quest: no VC, no board, no investors to answer to — they reinvest their own profits and ultimately answer to themselves. And however you read their motives, there's something refreshing about an AI company doing something that's legitimately trying to better humanity.

Next up: Runway opens a Sleep Clinic where you Dream in Latent Space?
(Ok, ok — still workshopping that joke…)

While not a "new" feature overall: LTX rolled out "Ingredients" for LTX 2.3 (their 22B open model). You know the deal — feed it a single reference sheet or multiple images, and it generates a video that keeps all of them visually consistent. Under the hood it's an IC-LoRA built on LTX-2.3, up on Hugging Face now, and since LTX-2.3 also runs on fal, you don't strictly need a local GPU rig to mess with it.

It's part of Lightricks' broader LTX Trainer release (June 17) — 13 training modes from a single YAML config (LoRA, IC-LoRA, full fine-tune), aimed squarely at the open-weights community. Ingredients is just the most immediately useful piece to fall out of it.

Why it matters: Ingredients/Omni-style modes are quickly becoming the go-to method for consistent AI video, so it's great to see it land on the open-source side.

My take: Haven't played with it yet — it just dropped this week — but I'll be digging in soon.

Luma widened access to Modify Video V2 in Ray 3.2 (which originally dropped June 9), and it's now available to everyone.

While I know that as a straight video generator, Ray-3x has not really blown up in the post-Seedance landscape, I will say that one of the secret superpowers of Ray is in its Video to Video capablities.

For this one: you can swap elements while keeping the original lighting and performance intact — at up to 20 seconds and 1080p, with frame-level control, improved motion transfer.

My Take: I’m still playing around with this, and while yes, on the surface this might appear to be a “Omni” type thing, it seems much more aimed at filmmaking rather than the “everything goes” approach of Google’s Omni model. Well, that and Omni taps out at 10 seconds.

The kind of crazy part is the 16 keyframes, which sounds kind of excessive and unwieldy, until you see it in action— and then it kind of makes sense.
I’ll dig into this one more soon.

Luma's Ray 3.2 announcement is here.

The distance between a client brief and a finished cut used to be measured in shoot days, crew, and reshoots. Artlist AI is built to compress that—a production tool aimed at professional filmmakers, agencies, and production teams who need to scale output without dropping below studio-grade. And AI Avatars are now live, joining lip sync and dubbing in one connected pipeline:

Because avatars, lip sync, and dubbing live in one workflow, the point isn’t “real footage or AI”—it’s both. Mix captured material with AI elements, ship fifteen tailored versions of a campaign instead of one, adapt a single piece for every market your client is in.

And with Unlimited now on select annual plans, you can do all of it without watching a credit balance—run as many versions as the work needs, not as many as your credits leave room for. It opens up work that wasn’t on the table before: the barrier stops being budget and logistics, and starts being your own taste.

If you’re running a real production pipeline—agency, studio, or in-house—it’s worth putting through its paces.

See it in action—try it on Artlist → HERE!

Kind of like Midjourney Medical, this is another one that seems pretty unexpected — but also just a ton of fun.

Google Earth rolled its Flight Simulator out to everyone on the web this week (alongside some pro additions like elevation profiles and new import types).

You can take off and fly over basically any spot on the planet, right in the browser, no install.

While I wouldn't call this a full sim a la Microsoft's Flight Simulator, which interestingly uses Bing Maps to populate its globe, it is neat to see Google just toss this out as a little experiment.

Again, this isn’t a MS Flight Sim competitor in the least, as Google mentions:

“The flight simulator is designed for casual exploration rather than high-fidelity aerodynamic training.”

So, while it is just Google Earth with a HUD, it’s also a LOT of fun, especially if you crank Danger Zone in your earbuds while exploring!

It’s a little weird to find, so per Google:
To launch the flight simulator, do the following:

Remember to fasten your seatbelts and have your tray in the upright position!

As of this publication: Fable is STILL blocked.

The Claude Fable 5 drama continues on. Earlier in the week I covered the basics— which has since turned into a moving target, with (as of this newsletter) speculation that it might return next week?
We’ll see…I’d rather NOT do a follow up to this video!
If you want my thoughts on WHY Fable is so good, check the video out here:
https://youtu.be/SlnzFb4psdc

The Week AI Video Got Cheap. The two biggest video models on the planet both quietly dropped budget tiers in the same stretch of days — Seedance 2.0 Mini at roughly half the price of full 2.0, and Kling v3's Turbo, which feels a LOT like Kling 3.0 at half the price!

Full run through, PLUS: A look at some Wild Real Time Video Models!
https://youtu.be/-oFqLNaKepg

Onward to next week!
Does Fable Return?
Will there be a surprise Video Model Drop?
Will GPT-5.6 Release?
Will Runway Open that Sleep Clinic?
Tune in next week to find out!

If you have any feedback or anything you’d like to see, please let me know: [email protected]

Or, just drop a comment on one of the videos! I honestly do see them all!
As Always I thank you for Reading…
Tim