What in the world. So there’s a big use of machine vision and even image generation in medicine, and Midjourney obviously does have some very smart people deeply familiar with the topic, but if they were going to contribute to this field, it would probably be with a real medical diagnostics company using some of Midjourney’s imaging expertise, not Midjourney creating what to be their own invention.
I’ve seen surprisingly no discussion on reddit or any of the tech blogs though I haven’t gone looking for anything specifically. I don’t know any technical details besides what’s on the page. It looks like sort of a CT version of an ultrasound, maybe. It’s… interesting. But real medical discoveries tend not to open as a “spa” and this looks like it’s aiming at a boutique pseudoscience audience rather than a real medical device. But it’s… interesting. And it’s not magical. Ultrasound imaging with advanced compositors are a plausible mechanism. I think it’s slightly plausible that they could’ve come up with something even though this is absolutely not where you expect medical innovations to come from. I should probably dig around tonight for analysis.
Also, there were some flaws with that script I posted earlier – trashing deletes from the last image you generated on down, not the one you’re looking at, sorry about that. The other functions work (mostly). It’s hard to pick apart get dynamically generated infinite scrolling content to hijack functions they should’ve added but didn’t. If anyone is interested in using it I’ll post a later version that tries to handle quirks as they pop up, but I’ve been having a difficult time with the trashing system, which is one of the systems that most needs a hotkey because they make you use several clicks to trash individual images (you can trash batches in one click).
So I did a little bit of digging and here’s my opinion
Not quite pseudoscience, but has a lot of hallmarks of pseudoscience. Like, it’s not magic. they’re not using crystal healing or quantum energy or anything like that. Ultrasounds are a real thing. The equipment they designed uses real parts that are capable of doing ultrasound imaging.
But… they’re opening a spa. And not a clinic. Their current position is using the FDA “make no claims about diagnosing or curing disease” health and wellness exception that pseudoscience/alt med relies on.
Reading between the lines, it’s going to be a very fancy version of those body fat / body composition readers that spas already have, which will look a lot cooler because it can probably image things like your fat distribution and not just a percentage. They’re hoping as they gather data they can come up with something that’s actually medically diagnostic and maybe apply for FDA status later.
So it’s weirdly in between real science / real medicine and pseudoscience and bullshit in a way that not many products are.
Their growth projections are absurd though and don’t seem at all like Midjourney. Like – midjourney is one of the few AI companies that aren’t using venture capital and losing money to gain market share. They have a traditional business model - you pay us, we give you a service, we’re profitable, we’re not running on venture capital. So then when they say they want to do a billion scans a year in 5 years, which is like… 5x more than all the MRIs in the world – that’s the sort of absurd growth numbers of the traditional “infinite growth venture capital” model, which is not very midjourney-like at all. The whole thing is super weird.
It’s not implausible. And it’s not crazy to think they can sort of image now, figure out what to do with that later – that’s the sort of thing AI analysis is actually pretty good at figuring out. It would be absolutely wild if a genuinely new useful medical imaging modality were created by a small, niche AI image generation company that until recently ran their business through a discord bot.
The thing to remember is that David is a big weirdo. I’ve listened to enough of his weekly Discord voice chat things to be in no way surprised that he’d be sincerely moon-eyed over some loopy medical application and his ability to make it happen. I once spent an hour listening to him talk to some dude about whether or not MJ users should be considered “mages” for their ability to create things. He also likes side projects such as his attempts to make a print journal of MJ art happen. This is obviously way larger and further from the core MJ wheelhouse but it feels (to me) more like “Well, that finally happened” than “How could this happen?”
Aha. I had no idea. Apparently it has been a sort of lifelong idea of his to invent some sort of medical imagining. And sometimes success in one field convinces you that you’re a genius that can have success in any field.
They could probably have a successful product if they just marketed it as a health trend instead of “this is gonna replace most medical scans for a tiny fraction of the cost and save millions of people.” There would be a huge market for people who could get pictures of, say, visceral fat and muscles every couple of months to track their fitness progress. People would love to post that shit on instagram, it would go viral, you could charge people $200-300 a pop and they’d gladly pay it at high end gyms and spas. And the product may be genuinely good for something like that. It could be genuinely useful in a way that most fitness trends are not. And that’s basically what they’re doing at their spa. Assuming it works well for what it sets out to do, like a very fancy DEXA scan that gives you cool images, that’s a real product.
So then why all the “we’re gonna revolution medicine” shit – I guess that’s where the founder’s personal dream comes in. It’s not like they have investors to impress, it’s a private company with no venture capital. That makes a weird but practical idea read as insane pseudoscience. I actually think their San Francisco Spa might be successful and expand, if they actually use it for what it makes sense to use it for. It almost feels like they want to be theranos even though they have a product that might genuinely be useful if more narrowly than they want, and, well, they don’t have any investors to scam.
Ah well. Back to the image generator for a moment - they have a cool feature where you can rank images for aesthetics and/or your own personal profile builder, and the top 2000 rankers per day get a free hour of GPU time. Which is great. Except you can’t preferentially spend it first, it only kicks in after you’re out of fast hours from your subscription. And my fast hours expire in about 2 weeks. Which means I need to burn the rest of my subscription’s gpu time to be able to spend those extra hours or I’ll lose them. They should really let you choose whether to burn the extra fast hours first or last.
If you’re on the $30 Standard plan, the real benefit is that you can set it to Relaxed and go well over your paid thirty hours with the free relaxed time. The possible bonus hours for rating is nice but only if you’re really using MJ a lot.
If you don’t use MJ a ton, it might be worthwhile to drop to the $10 tier for ten hours a month and no Relaxed option. Then you just rate images for a free hour – I’ve never NOT gotten my free hour if I was trying for it – and keep the pot full that way assuming you don’t suddenly got crazy on inspiration for a week. In theory, you could get 30 free hours a month that way but, if you’re gonna make a job of it, just pay the $30. If you just do it while watching TV or kicking around then the supplemental hours can still make a $10 sub feel flush.
That’s actually a good point. I rarely even empty the regular GPU fast time to need relaxed - especially since the new model has brought down generation costs. I got the free hours mostly because they give you credit for building up your own profile. I just got frustrated that I basically had to spend all my regular fast GPU hours to use them before they expire.
Does the $10 plan have ten hours a month? It just says “200 image jobs per month” which seems like less than 10 hours for me. It’s a little cagey that they won’t just list the GPU time. I don’t mind paying the $30, but if the difference is only 5 hours and I don’t use the relax generation anyway, maybe I should see if the $10 is adequate.
Sorry, 200 minutes. So that’s 3.3 hours but I personally rarely stress it. Not sure where 10hrs popped into my head from. Mine refreshed recently and I haven’t used it since so here’s my /info
Subscription: Basic (Active monthly, renews next on July 18, 2026 9:08 AM)
Job Mode: Fast
Visibility Mode: Public
Fast Time Remaining: 200.00/200.0 minutes (100.00%)
Ah yeah that makes more sense. I’m pretty happy with the $30 tier. Midjourney being one of the few companies not to give you subsidized service to grow market share like pretty much every AI company makes me feel good about doing business with them. They’re giving you an honest deal. That philosophy of having a traditional business model and being responsible and responsive to your customers is part of what makes the medical imaging business such a wild left turn.
And I do use it a few times a month and very much enjoy the images and workflow even though I have access to multiple other image generation systems. I just rarely exceed the 15 hours I’m given, sometimes I’ll make a mad dash at the end of the month to generate a bunch of videos for images I like as to not lose/waste the remaining GPU time.
Maybe I’m being too stingy with my generations and I should be generating 5 or batches where I’m currently generating 1-3. I do usually do 3 at a time now, though, through various permutations like -stylize {0, 200, 500} or the multi-model -v {7, 8.1, 8.1 --raw} or -weird {0, 20, 50}. I tend to find spreading out the idea over 3 generations gives interesting results.
Interesting and versatile new feature. They added a draft mode to version 8.1, hit the lightning symbol to the right of the text prompt. Enter a prompt and you get 24 low res takes on your prompt. They’re all far too low res to be usable but they’re high enough to explore the concept quickly at almost no GPU cost (apparently all 24 cost half as much as one 8.1 standard def batch).
Here’s an example output. This is “a post apocalyptic world --chaos 50” with a 4:5 aspect ratio and stylize 100.
24 candidates with very different takes is an incredible amount of ideas to explore for almost no GPU time and very quickly. I never used the draft mode with v 7 but this is genuinely different - it’s not just a cost saver, it’s a quick way to explore many takes on a scene. On any one of these scenes you can run vary subtle, vary strong, remix, and it’s like doing a creative upscale - preserving the vibe of the image while changing and filling in the details suitable to the higher res image.
Okay, after a day of use, the v8.1 draft mode is fantastic. That’s exactly how I like to use midjourney, only condensed. I usually run multiple batches of any idea with medium chaos to feel out what sort of interpretations of the idea are available, and the draft mode pretty much does exactly that only faster and at a tiny fraction of the cost. I could see this becoming my main midjourney workflow - generate a draft sequence, figure out which pics were good ideas, and then rerun them at full res (usually with “vary strong” but if I really like the original image, then “vary subtle”)
The news I saw on the midjourney discord wasn’t actually draft mode - that was released 10 days ago - but another feature they came out with. --sref random. style references are pre-built styles you can use to affect the color, type (animated vs painted vs photorealistic, etc), type of characters, and general vibe of the pictures. You can combine draft mode and sref random to get a very wide examination of an idea.
For example, here’s “A dirtbike rider tears up the terrain” with --sref random
I could’ve come up with a better prompt but this demonstrates what I mean. The random styles are all over the place, photorealistic, black and white sketch, colorful psychedelic, all sorts of illustration styles. A wildly creative variety of ways to interpret that prompt practically for free. And you can iterate on any one of them.
And if you find a style you like, the sref number is attached to the picture so you can use it in the future.
