I asked AI to generate an image of the Andromeda Galaxy with specific capture parameters: a 300mm apochromatic lens, 1-minute subs, and 300 minutes of total integration time.
When it popped up, it looked terrifyingly accurate.
 |
| AI generated image of Andromeda Galaxy with specific prompts of lens, integration time and subs |
To test it, I fed the picture into a couple of other AI models. Unsurprisingly, they all identified it instantly—pointing out the dust lanes, the blazing core, and even satellite galaxies M32 and M110. One AI went as far as guessing the exact hardware setup used to shoot it—the very prompt I had fed the first AI.
My stomach churned. Was this where the hobby was headed? Smart scopes were already doing the heavy lifting—locating, tracking, and stack-processing on the fly. Was authentic astrophotography about to be completely rendered obsolete by synthetic pixels?
Then I stopped and looked closer. Something felt off!
The star field didn't look right. Though nobody memorizes every background star in the sky (at least not me 😊), after staring at countless exposures over the years you develop a feel for the neighborhood. The stars in this image seemed randomly sprinkled rather than what my partly experienced eye could tell. I still consider myself a beginner to intermediate stage in the hobby.
Enter 'plate solving'.
For anyone unfamiliar, plate solving is how astrophotographers verify where a scope is pointed. You do this before any imaging session when you don't have a GoTo capability to frame your celestial object. You feed an image into a database that cross-references the precise geometric patterns of stars against mapped catalogs to extract exact celestial coordinates. If an image is real, the stars must align.
I uploaded the AI image to a plate solver. The site groaned, churning through star catalogs for what felt like an eternity.
While waiting, I ran a control test. I fed the solver one of my old raw captures—it identified the field instantly. I tried a verified image from the web—solved in seconds.
Finally, the AI image result came back: Failed.
 |
| Results from a plate solved test of an uploaded AI generated astroimage online |
I heaved a massive sigh of relief. The AI could fake the visual aesthetic of deep sky objects, but it couldn't fake the immutable geometry of the universe. Hard data and human intuition won the day—at least for now.
