AI summary by TwinMind
How Samsung Phones are "Faking" their Photos.
Mrwhosetheboss
10m 33s
Samsung Moon Photo Claims and Initial Investigation
- Samsung advertises phones with up to 100x space zoom capabilities for years
- Users can zoom into moon, take shot, get results crisper than $5,000 professional camera setup
- Reddit user "I break photos" downloaded high-res moon image from internet
- Downscaled image to 170x170 pixels (tiny resolution removing all detail)
- Applied blur to manufacture low-detail image phone might identify as legitimate moonshot
- Turned off lights, used Samsung phone to photograph blurry 170x170 moonshot
- Result showed incredible difference - concluded Samsung using AI Scene Optimizer to place craters where original was just blurry mess
- Claims phone not just sharpening original image but recognizing moon and swapping with correct texture
Personal Testing - Four Photo Experiments
Test 1: Genuine Real-Life Moon Shot
- Captured actual moon at night
- Resulted in impressive looking image, nothing unusual
Test 2: Photo of Moon on Screen
- Used photo of moon on MacBook screen placed on dark sky scene
- Phone not super easy to fool - knows you're never sitting right next to moon
- 1x main camera just looks like photo on screen
- After 5 minutes getting right distance for 25x zoom, scene detection kicked in
- Recognized moon, transformed shot significantly
- Expected quality loss from 25x zoom of photo didn't occur - implies "fiddling"
- Control test: Cat's face in same starry sky background showed noticeable quality degradation
- Something specifically happening with moon detection
Test 3: Recreating Reddit Experiment
- Took original photo, downscaled from 2000x2000 to 200x200 to remove details
- Added blur so no visible moon texture remained
- Samsung produced shot with zero detail loss from source image, actually had more detail
- Phone saw only white blob but knew dark patch was crater, blurry bit was moon
- Only explanation: phone recognizing moon shape, using pre-existing knowledge of moon appearance
- Printed blurry moon version, hung from light by string - phone detected paper as moon, added non-existent detail
- iPhone 14 Pro: no extra detail added
- Google Pixel 7 Pro: photo remained blur as expected
Test 4: Complete Deception Attempts
- Tried lighting ping pong ball in black room - didn't work
- Tried suspending brightly lit blank paper - didn't work (but phone removed hanging wire)
- Samsung good at differentiating actual moon photos from brightly lit white objects
Definitive Testing Method
- Took blurred moonshot Samsung recognizes, scrubbed out detail, replaced with own detail
- If Samsung just overwrites with stock moon image, tweaks would disappear
- Result: Phone recognized moon, added detail but kept tweaks made
- Proved using actual photo taken as base
- Phone identified added image as moon crater, fused images, applied crater texture
- Samsung trained AI to understand real moon texture using hundreds of thousands of real images
- AI recognizes moon presence in future shots when enough identifiers present
- Uses learned knowledge to analyze each moon photo from scratch
- Recognizes familiar bits, adds those in, improvises enhancement for rest
Technical Analysis
- Beyond normal tweaking phones do for random objects never seen before
- Better than accusations of using pre-baked image slapped on top
- No pre-baked image found in camera app files
- Moon view locked from Earth - AI only had to learn basically flat JPEG image
Concerns and Controversies
Moral/Philosophical Concerns
- Company doing this without being open about it
- Changing sanctity of what means to be photo vs generated image
- AI age where anything can be generated - do you want photos/memories to become AI-generated?
- Counterargument: Smartphones already do much useful processing that wouldn't be possible without software interference
- Phone cameras would be stuck in last decade quality without software enhancement
- Phones already recognize text in distant signposts, enhance legibility using text knowledge
- Author prefers sharper text and richer detailed photos over "genuine but bad-looking" shots
Marketing Deception Concerns
- Samsung used moon photos in marketing multiple occasions
- Claims phones have such good cameras they can zoom 100x into interplanetary object with crisp output
- Marketing trickles through to comparisons and organic content
- People test phone side-by-side with others, conclude Samsung cameras "destroyed everyone else"
- Using AI specifically trained for one use case to look incredible
- Shooting anything else results in normal performance
- Can turn extremely blurry shot into perfect moon with software trick
- Could substitute potato laptop webcam and still get perfect moon result beating 90% of other phones
- Author says this isn't really great camera performance
Samsung's Response and Market Impact
- Samsung should have communicated better what they're doing
- Hidden moon mode monopolized market for moon photos due to marketing
- Should have avoided linking moon photo output to camera quality claims
- Feature called "Scene Optimizer" indicates AI enhancement
- Users can turn off feature for more genuine photos
- Debate: Would users rather have blurry authentic shot or enhanced detailed shot?
- Author argues if eyes can see moon, AI allowing phone to do same captures memory closer to how remembered
- Process less authentic but end photo arguably more authentic
Industry Context
- Smartphones already use software to enhance images in many ways
- Make grass greener, skies richer, tweak facial features
- Unsolicited photo tampering for faces more invasive than moon adjustment
- Tech news sites extensively covering this controversy
- Samsung critics see as proof of deceptive practices
