The stars were stripped out with neural network tools (StarNet++/StarXTerminator) at the studio request so text credits would read cleanly over them. The underlying nebula data is real, but removing every star from the field puts this firmly in the category of art photography, not scientific imaging.
No one has ever or could ever observe a nebula with zero stars in the frame.
That is not what starnet does. It just removes the star from the picture you took, nothing else. It also predates generative AI by a few years.
If by "not real", you mean "you removed the stars so it no longer reflect reality!", then real photograph doesn't exist. For example, OP uses narrow-band filters, and it's common to map H-alpha wavelength, which is red, to green in the images. Does that make it unreal?
In the end, astrophotography is more art than science; the goal is more about producing aesthetically-pleasing images than doing photometry. Photographers must take some artistic license.
“StarNet is a neural network that can remove stars from images in one simple step leaving only the background. More technically, it is a convolutional residual net with encoder-decoder architecture and with L1, Adversarial and Perceptual losses.”
Your citation is a valid copy-paste from the website linked, but you haven't yet replied to the assertions of the parent comment. Could you, in your own words rather than someone else's, speak more about your concerns and address those assertions?
They're a lot more real than CG/AI. It's very rare and maybe not even possible to have a "true" astrophotography photo.
At those light levels, eyes and camera sensors work very differently and even a "plain" astro photo has either been processed a lot, or else doesn't look like what our eyes would see.
Even straight-out-of-the-camera JPG files have been heavily processed - they are just hidden behind the RAW processor which we have taken for granted; not to mention smartphone photographs, which employ neural network in the processing pipeline.
So this part of the blog post is essentially false: "no generative AI of any kind"
I have yet to see a precise technical definition of what "generative AI" means, but StarXTerminator uses a neural network that generates new data to fill in the gaps where non-stellar objects are obscured by stars. And it advertises itself as "AI powered".
I don't consider photos I take on iPhone to be "AI generated" or even "AI augmented" even though iPhone uses neural networks and "AI" to do basic stuff like low light photography, blurring backgrounds, etc.
I agree that I wouldn't call these photos "AI generated", because the majority of what you're seeing is real.
But that's very different to saying that no generative AI was used at all in their production. "AI augmented" sounds pretty accurate to me.
Likewise, if someone posted a photo taken with their iPhone where they had used the built-in AI features to (for instance) remove people or objects, and then they claimed that no AI was involved, I would consider that misleading, even if the photo accurately depicts a real scene in other respects.
As a photographer and machine learning guy, I would call a lot of modern phone photos AI augmented. AI to stack photos or figure out what counts as the background is a little bit of a gray area, but an img-to-img CNN is about as close as you can get to full AI generation without a full GAN or diffusion model.
“StarNet is a neural network that can remove stars from images in one simple step leaving only the background. More technically, it is a convolutional residual net with encoder-decoder architecture and with L1, Adversarial and Perceptual losses.”
> So these are more artistic photo works than real science photos...
I disagree. If there are many flies around a statue, and I photograph the statue but remove the flies in the photo (via AI or any other technique), then I'm still producing an image of something that exists in the world - exactly as it appears in the world.
I agree that the claim "no generative AI used" is technically incorrect, but I do feel that the image does not contain any AI-hallucinated content and therefore is an accurate representation of reality. These structures appear in the image exactly as they exist in nature.
AI-related definitions aside, if it's a strictly subtractive/destructive tool that only removes light, it's hard to characterise as "generative" and arguably not much different to filtering frequencies!
It's not just "removing light", because if you removed all the light from stars, you would be left with black spots instead of white spots. The stars are bright enough to completely saturate a region of the image sensor. So there was actually no data recorded about what was in that particular part of the nebula or whatever.
The "generative" part is that the algorithm is filling in a plausible guess as to what would have been observed if there was no star "in the way".
Thanks! and yes, that's the summary!.
The distribution matters too. GPT-4o at 10.6% vs Gemini at 56.1% is a 5x gap between first and last. And the highest-bypass category across all five models was social engineering / identity impersonation at 35%, which maps directly to the indirect prompt injection problem in agentic deployments.
The fact your work is independent of the vendors is a major plus. My recommendation is to continue to develop, refuse any "colaboration" with these well funded companies.
I could see this turning into a valuable third party resource, you can even monetize, for companies implementing agentic solutions. The industry needs independent third party voices.
That's exactly the intent, independent, reproducible and no vendor relationship.
The monetization angle is interesting. A continuously updated version with more models and frontier models, agentic scenarios, and multi-turn testing would be genuinely useful for teams making deployment decisions. That's the direction for v2.
I wonder what the average career tenure of the userbase here is now, because Github was slow and flaky well before Microsoft got involved.
Maybe it wasn't as noticeable when Github had less features, but our CI runners and other automation using the API a decade ago always had weekly issues caused by Github being down/degraded.
"Funny" enough if this controller had had a medical emergency (or just bad sushi) and been off the radios, this wouldn't have happened because the fire truck would not have received clearance to cross the runway and wouldn't have. Or at least would have crossed like the airport was uncontrolled, been much more careful and announced itself, and likely have seen the landing aircraft.
And if an aircraft needs to land due to an emergency? It’s amazing things work as well as they do, the system relies on only one thing going wrong at a time. This accident was an example of multiple things going wrong at the same time.
Every accident is an example of multiple things going wrong at the same time.[0]
I'm going to pretend to know exactly what would happen in that precise scenario but I'm confident most commercial pilots get enough training to be able to handle it.
>> Every accident is an example of multiple things going wrong at the same time.
You are defeating your own argument :-) Its exactly because every accident is an example of multiple things going wrong at the same time...that you need...multiple layers of control and safety to catch it through each hole of the cheese.
One of the things you learn as a pilot is how to recognize that you need to go into emergency mode if you will. Call it high-alert if you want.
You need to recognize when something is out of the ordinary and treat it as an emergency (perhaps not a literal pan-pan/mayday emergency) sooner rather than later, and do things that may end up to have been unnecessary (like executing a go-around because emergency vehicles were on the move).
One controller on two frequencies is another example - that works fine in normal situations, but during an emergency response, perhaps the channels should be mixed; giving the pilots in the air a chance to hear the incorrect clearance onto their runway.
After all, an active runway is really more of an "air" control thing than a ground one.
An empty tower at La Guardia with a bunch of airplanes in the air not getting a reply to their calls is Die Hard 2 stuff. Spare me the Pete Hegseth school of ATC...
I have absolutely no idea what you're talking about. The GP is literally about a lone controller in the tower having a medical episode and what would happen after that.
The pilots would execute untowered approach procedures, a small airport with little to no traffic and VFR flight you may self-announce on frequency, a larger airport you go back to approach, etc.
Each of those flights should have an alternate and be prepared (have enough fuel) to divert. If there is a fuel emergency then self-announcing is likely appropriate as the plane is coming down anyway, but that is multiple things going wrong.
A big part of it is what category of airport it is, and plane. General aviation almost always goes to self-announce (which includes some business jets perhaps, they often land at untowered airports) but not category 135 air travel or whatever it is.
You should try harder, because I'm not making any comment on regulation whatsoever. There are procedures that every controller and pilot knows for how to handle loss of radio contact.
Just a quick read/speculation based on the linked forum post...
Short of insane visibility conditions that prevented them from seeing the plane coming, the firetruck operator seems to be the liable party (beyond the airport for understaffing controllers—this seems to be exacerbated by government cuts but that's still no excuse for having a solo controller at that busy of an airport, especially at night).
The controller in question seems to have caught their mistake quickly and reversed the order instead asking the firetruck to stop (but for some reason, this wasn't heard).
Is it common now to have solo operators running control towers?
"Liability" isn't really how we try to see things in aviation. While it's true that it's ultimately considered the responsibility of the truck/plane to visually confirm that crossing the runway is safe, refuse unsafe commands from ATC, and comply to the best of their ability when ATC says "stop" at the last second, we can't stop our analysis there if we want to prevent this from happening in the future, because unless things change someone will make this mistake again in the future. Telling people not to make mistakes isn't going to help at all; it's obvious, and no one wants to cause an accident. The error is just the last step in the process that led to the collision.
I don't think the ATC is at fault here. If they were put in a difficult situation and responsible for too much at once, I'd view that as a leadership bug, not their personal fault (or anything they should be held liable for). The weak links imo here are the firetruck driver and whoever that ATC reports to directly (i.e., there shouldn't have been an opportunity for this to happen—that's an executive failure, whether they want to take ownership or not).
The weak link is the system in place which puts so much work on so few staff.
The fire truck received the go ahead. They weigh 3x more than a normal firetruck. They're rushing to a different emergency. The plane is moving fast as hell. They can't just react instantaneously.
The ATC worker is clearly too stretched and such an incident was an inevitability. When they're shouting stop, they are no longer directly talking to the firetruck, which obscures the situation for everyone.
It is a terrible tragedy that will only be prevented with reform in staffing and safety procedures.
We aren’t in the aviation industry, and neither are we the NTSB.
Prosecutions and convictions do occur as a result of aviation incidents, pilots loose their jobs, pilots loose their licenses, ATC staff can be prohibited from ever working in the industry again.
The controller was talking to Frontier plane when he first said stop, then said stopstopTruck1stopstopstop and it would be easy for there to be a gap in processing for the driver of truck 1 because the verbiage all flowed in the same stanza that was started when addressing the Frontier flight.
Not arguing with the regulations, just pointing out that based on airport diagram[1], since the truck was crossing rwy on taxiway D, the CRJ was on the right approaching from behind. I have never been inside an airport firetruck, but I guess from the driver's seat the jet would be quite hard to see in this case.
"While driving on an aerodrome : Clear left, ahead, above and right
Scan the full length of the runway and the approaches for possible landing aircraft before entering or crossing any runway, even if you have received a clearance."
He was stopped until he received instructions to cross the runway from the person whose job it is to sit in a position with good visibility and tell people when they can cross runways. He wasn’t driving fast at all. The whole system is set up so that vehicles with blind spots (every large passenger jet) can safely move.
We can’t say that emergency vehicles should just stay in on dark and stormy nights.
>from the driver's seat the jet would be quite hard to see in this case.
..is what I was responding to.
>We can’t say that emergency vehicles should just stay in on dark and stormy nights.
This conclusion is flawed and doesn't apply to what I said.
If a truck can't see (conditions or not), then they shouldn't be on the same runway as takeoff/landing because...the consequences were severe despite the safeguards you mentioned, e.g. Not driving fast is relative and the "eyes" failed too initially.
“Vehicles with large blind spots don’t belong on the runway” is a completely untenable proposal.
Almost every airplane is bigger, blinder and slower than that truck. If it had been a plane cleared across the runway, this would have been so much worse.
Even if you want to exempt airplanes, it would require a complete rebuild of most major airports or using completely different emergency equipment. Every airport you have ever flown to commercially has ground vehicles crossing or operating on runways every day. It is simply not possible to operate a commercial airport without ground vehicles in aircraft movement areas, including runways.
The solution is not to spend billions on new trucks or access roads because of a single incident. It is to ensure that controllers, the people directly in charge of coordinating safe ground movement, have the mental bandwidth and tools to do their jobs. The fact that this was a truck and not an airplane is luck, making any discussions about truck cab visibility very much secondary. You have to go upstream of “trucks have blind spots” to truly prevent another of these incidents.
You even earlier: “ But if your truck has blind spots and vis is poor, you shouldn't be driving as fast if at all”
How do you propose that a truck not driving “at all” manage to drive on the runway? Driving on the runway, (or anywhere) is a subset of driving “at all”. Logically I can conclude that since you think that the trucks should not be driving “at all” due to blind spots, that you also think that they should not be driving on runways because of blind spots.
My argument paraphrased you to highlight a specific situation that would arise as a result of what you argued and to point out the folly of just banning any vehicle with a blind spot from crossing the runway. By extension, that planes can’t cross the runway either (the difference between a fire truck and an airplane crossing the runway is that the plane is larger, with bigger blind spots, less maneuverable, fragile and filled with people).
The solution is not to ban vehicles with blind spots from crossing runways, but to provide tools and guidance for those vehicles to operate safely. You could, for example, provide them with a trained observer in an elevated place that can be responsible for saying whether it is ok to be on the runway. We could give the person coordinating movement in the elevated place tools like radar mapping the ground, or automated semaphore systems at runway crossings (I’m describing things that already exist). Using a system like that we could do things like operate in 0 visibility where the weather causes the blind spot to be anything past the windshield (which is something that happens at JFK for example).
They were all, including truck 1, queued up at the stop line waiting for clearance to cross. Truck 1 received clearance to cross, he began crossing, then received instructions to stop after it was too late.
The rest of the emergency vehicles were stopped because they hadn’t been authorized. Truck 1 started moving because he had received specific instructions to do exactly what he was doing.
I take it you’re not a pilot, controller or someone who has ever worked an aviation radio?
"Truck 1 and company" were cleared to cross. A few seconds later, "truck 1" was instructed to stop.
Edit: Confirmed truck 1 was the one involved in the collision. Previous text: It is unclear which truck specifically was involved in the crash. In photos, the truck has the number 35 on it, not sure if that would preclude it from being identified as "truck 1" verbally.
Ah. I missed hearing that “and company” in the recording.
In any case, if they were cleared across the runway, and they were, it isn’t really on them. It doesn’t change the gist of the argument. The broader point is that it wasn’t that one truck was barreling around being reckless as implied by gp, it’s just that one truck made it out and the rest of the company had yet to start moving (whether because they saw the plane coming from their viewpoint farther back, or just hadn’t started moving yet, we will find out later). The entire company had stopped at the line, and when cleared across the lead truck was struck. Of course the rest were still stopped behind the line, there was a giant fire truck in their path moments before.
The instruction to stop is, to my pilots ear, irrelevant. Until an instruction is read back by the receiving party, it is worthless. It might not have been received, or received incorrectly. That’s the whole point of the readback, to ensure that the instruction was received correctly (notice how I missed the “and company”… a readback would have caught that). If there is not a readback, controllers are instructed to ask for one. On top of that, it was a panic instruction using non standard verbiage. If he was already past the line, the instruction to stop might have made it worse.
“I take it you haven’t ever worked with radio.. “. Seems like you haven’t a clue how any of this works. Doesn’t matter if they had radio clearance, the fire truck is responsible for ensuring runways is clear and not driving in front of plane.
I’m a certificated pilot in two countries, trained in this region, and own an airplane. I have a pretty good grasp how this works, but am willing to learn if you have citations besides the CFR pull quotes elsewhere in this thread.
All people (pilots included) are responsible for only following ATC instructions if it is safe/possible to do so. You aren’t supposed to land on a runway with other traffic on it, even if cleared. You aren’t supposed to cross a runway if there is a plane taking off or landing, even if cleared. You aren’t supposed to clear a vehicle onto a runway at the same time you cleared a plane to land (this one’s an assumption, I’m not a trained controller).
You are making the assumption that the truck did not check the runway, but keep in mind that it is a 30ish ton vehicle, and the plane was moving at 150 mph at touchdown, 100 mph at the time of impact. There very well may not have been a plane visible when the truck started moving. The truck might not have received the non-standard clearance revocation, or received it and tried to get off the runway by accelerating across, or received it and begun slowing in the path of the plane.
The truck driver could have prevented this, but they certainly aren’t the primary cause.
I very much doubt that you know the exact timing of the event. My guess is that you might have seen a video where some industrious editor put the ATC recordings over the leaked surveilance footage, but there is no way that is correctly synced.
With publicly available information we can sync it to within ~2 seconds. All trucks other than the first one were definitely in the process of stopping in between the first and second time ATC told them to stop (5 seconds apart).
I've seen the NTSB footage of the plane and helicopter crash in Washington. It is practically impossible to discern a landing plane over the lights of a big city at night. Next, the truck had been angled away from the plane approach, so it was coming from the right side (passenger side) and at back. There was zero chance that firetruck could have seen the plane in the seconds it covered hundreds of meters, flying at 200+ km/h. And also in that NTSB investigation there was a case of missed comms, when ATC recording clearly showed controller saying an important word, but it jot "jammed" in the process and there was silence at the receiving end instead of that word (in the middle of the transmission). Not saying it was a case here, but it is possible too.
Just like in that collision, it is possible there is no one single person to blame (apparently helicopter pilot was not outside of the legal corridor, despite the speculations), but it was a compounding error issue.
Truck was on a different frequency from the aircraft so they couldn’t even hear each others’ clearances.
Also first time ATC told the truck to stop it wasn’t too clear who the message was addressed to. It’s a bit hard to hear “Truck1” there, not clear who he wants to stop. The second time, one can argue by the time “stop” command was heard it might have been better to gun the engine. As the truck sort of slowed down in the middle of the runway.
> The crash has raised fears that operations at US airports are under extreme stress. Airports have been dealing with a shortage of air traffic controllers, exacerbated by brutal federal government personnel cuts by Donald Trump’s administration at the start of his second presidency.
So where there budget cuts or not? That was the claim. I have yet to find anything that suggests there were budget cuts, just vague mentions of "brutal federal government personnel cuts".
I'm just looking for: budget was X in <2026 and in 2026 it is Y, where X > Y
I made a face, reading this article. They present the US gov't's very large and scary liabilities and future obligations, but they don't present the other side of the picture, the future income streams. (How much can the US government realistically expect to earn annually via taxation?)
Without being able to compare future liabilities to future income, we're lacking critical context. It's like they wrote half an article; kinda frustrating.
There is no feasible scenario where tax revenues will allow the US government to pay a 39 Trillion, soon to be 40 Trillion debt. And paying the debt its not even in discussion right now.
What is in discussion, are the multiple, very feasible, and very realistic scenarios, where an increase in interest rates, and a run from the dollar...Will force the US government to spend over 80% of the tax revenue, JUST TO SERVICE the debt interest....
I am not an economist but my worry is that government deficit spending was the largest driving factor for the bull run. Balance the budget and the economy crashes.
No one has ever or could ever observe a nebula with zero stars in the frame.
reply