PLA next/6th generation fighter thread
- Thread starter Blitzo
- Start date
- Status
The funny thing is, I wouldn't be surprised if indeed "something happened" today but just because these photos are fake, everyone is now very careful at the moment a) to post something and b) to judge So it will be quiet again until maybe a little later then real photos appear.
Last edited:
No peeking of gifts before the 25th!
Speaking of which, instead of an early Christmas, perhaps the Chengdu AC's alleged/purported J-XD's first flight should be an early Chinese New Year celebration instead, considering it's supposed to be a monumental red packet for China.
Last edited:
This is a fake and a lazy one at that.
Here's a quick composite I made in Photoshop which takes this new fake image and a past real image from J-35 first flight or something where it was accompanied by a Flanker. I took the Flanker for reference because the colours match better than the light and bright colours of the J-35. The image is from SDF so you probably can guess which one it was. The top images are not modified. The bottom images have 200% contrast. Once you increase contrast on a compressed image the algorithm for compression is being emphasised. You can zoom in to see details better but some things should be visible already.
Compare first the sky i the real image has noise visible - the sky is all kinds of different shades in a random pattern just as you would expect in nature. The fake image has fields of uniform colour indicating a rendered environment. It has blotches of different shade which look like crude manipulation so it is likely a primitive composite image.
The plane is also modified and pasted in. The "halo" effect which results from averaging of dark and light colours by the algorithm for compression is not mathematically plausible. Pay attention how if follows the vertical stabilisers of the flanker or other dark parts. It is very even and uniform - because it follows a very precise dark shape against lighter background. The fake image has it looking as if it was drawn by hand by someone who draws poorly. It also disappears where the colour of the plane is not distinct from the background. The "halo" on the fake image disappears in some parts of the plane where it should be visible because all of the plane is very dark. This means that this is a manipulated element that was added to the background and in some places it had a few pixels of the old background left and in others it was cut or erased somewhere on the body of the plane so once it was pasted in, there were no "sky" pixels to work with hence the halo is visible in some parts and not in others.
Also this plane keeps no perspective whatsoever. It looks like a Flanker body that was modified but not in 3d. Some parts of it go one way, the other part goes another way. Someone said this is a screen from a game. It is very plausible. Uniform rendered sky plus a rendered plane model that was modified very crudely in 2d raster program to look "futuristic".
I thought people who watch for photos like these would learn a thing or two about photograph manipulation. It would save you so much uncertainty. It's very useful and not very difficult once you get the hang of the technical stuff i.e. that pixels have mathematical values for colours and that all algorithms just play with numbers in different ways. Just get yourself a simple free photo editor and read up on how the compression and other tricks work like layers and manipulating a raster image in 2d (stretching, distorting etc) and you will be able to solve 99% of fakes in no time. This is not difficult stuff but most people never think about it because we somehow tend to think that all images online are "real". They're not. Only analogue photos are real. Very high resolution digital photos with hig colour depth are approximations of real. Everything else including all PNGs, JPGs, WEBPs and whatnot are "drawings from photos" made by a number algorithm.
Most fakes are intended to excite people and cause an emotional reaction. This is why they are not done meticulously. Which is why you do the "dancing bridge" tricks that is amplify various parameters of the image several times to see where the errors or omissions can be. Just like the engineers do with bridges by amplifying deformations by 10x or 100x to see intuitively where the structure is being affected. In reality the bridge deforms by centimeters but on the screen of the computer it is bent out of shape by meters. And that's how you know what is being deformed rather than by how much, because that's the important bit. Whether there is a fracture or deformation. Because with bridges as soon as one forms the bridge falls. You never get to large deformations or fractures.
And that's how you test for fakes. Images - amplify contrast or colour palette (turn on just the blues and amplify 100% then reds, then greens to see where the colours don't match - a real photo doesn't have mismatches) or test contrast like here.
For movies - do slow or fast movent and watch for smudges or jumps. Even the best AI can't do it well yet because it's a whole new thing with numbers not working right. You're supposed to fool an impatient and excited human eye that is not looking to closely. You want that eye to see a shape and let the brain do the rest of imagining what it sees.
For still images or clips from video games or other simulated environments watch for backgrounds - noise in the sky. Uniform backgrounds save compute. The more variety the more compute is spent. Same with textures - they are being mapped over objects. if the same pattern is repeated you're seeing a render not a natural image. And you find repetitions by simplifying the image i.e. contrast which does the number trick explained above.
Remember - you are very rarely looking at an image. Your brain is matching the image on the retina with something that it already has in its memory. Even complex images work like that. The whole thing about learning drawing, painting or good photography is to unlearn that habit and just see shapes, colours, contrasts etc.
Here's a quick composite I made in Photoshop which takes this new fake image and a past real image from J-35 first flight or something where it was accompanied by a Flanker. I took the Flanker for reference because the colours match better than the light and bright colours of the J-35. The image is from SDF so you probably can guess which one it was. The top images are not modified. The bottom images have 200% contrast. Once you increase contrast on a compressed image the algorithm for compression is being emphasised. You can zoom in to see details better but some things should be visible already.
Compare first the sky i the real image has noise visible - the sky is all kinds of different shades in a random pattern just as you would expect in nature. The fake image has fields of uniform colour indicating a rendered environment. It has blotches of different shade which look like crude manipulation so it is likely a primitive composite image.
The plane is also modified and pasted in. The "halo" effect which results from averaging of dark and light colours by the algorithm for compression is not mathematically plausible. Pay attention how if follows the vertical stabilisers of the flanker or other dark parts. It is very even and uniform - because it follows a very precise dark shape against lighter background. The fake image has it looking as if it was drawn by hand by someone who draws poorly. It also disappears where the colour of the plane is not distinct from the background. The "halo" on the fake image disappears in some parts of the plane where it should be visible because all of the plane is very dark. This means that this is a manipulated element that was added to the background and in some places it had a few pixels of the old background left and in others it was cut or erased somewhere on the body of the plane so once it was pasted in, there were no "sky" pixels to work with hence the halo is visible in some parts and not in others.
Also this plane keeps no perspective whatsoever. It looks like a Flanker body that was modified but not in 3d. Some parts of it go one way, the other part goes another way. Someone said this is a screen from a game. It is very plausible. Uniform rendered sky plus a rendered plane model that was modified very crudely in 2d raster program to look "futuristic".
I thought people who watch for photos like these would learn a thing or two about photograph manipulation. It would save you so much uncertainty. It's very useful and not very difficult once you get the hang of the technical stuff i.e. that pixels have mathematical values for colours and that all algorithms just play with numbers in different ways. Just get yourself a simple free photo editor and read up on how the compression and other tricks work like layers and manipulating a raster image in 2d (stretching, distorting etc) and you will be able to solve 99% of fakes in no time. This is not difficult stuff but most people never think about it because we somehow tend to think that all images online are "real". They're not. Only analogue photos are real. Very high resolution digital photos with hig colour depth are approximations of real. Everything else including all PNGs, JPGs, WEBPs and whatnot are "drawings from photos" made by a number algorithm.
Most fakes are intended to excite people and cause an emotional reaction. This is why they are not done meticulously. Which is why you do the "dancing bridge" tricks that is amplify various parameters of the image several times to see where the errors or omissions can be. Just like the engineers do with bridges by amplifying deformations by 10x or 100x to see intuitively where the structure is being affected. In reality the bridge deforms by centimeters but on the screen of the computer it is bent out of shape by meters. And that's how you know what is being deformed rather than by how much, because that's the important bit. Whether there is a fracture or deformation. Because with bridges as soon as one forms the bridge falls. You never get to large deformations or fractures.
And that's how you test for fakes. Images - amplify contrast or colour palette (turn on just the blues and amplify 100% then reds, then greens to see where the colours don't match - a real photo doesn't have mismatches) or test contrast like here.
For movies - do slow or fast movent and watch for smudges or jumps. Even the best AI can't do it well yet because it's a whole new thing with numbers not working right. You're supposed to fool an impatient and excited human eye that is not looking to closely. You want that eye to see a shape and let the brain do the rest of imagining what it sees.
For still images or clips from video games or other simulated environments watch for backgrounds - noise in the sky. Uniform backgrounds save compute. The more variety the more compute is spent. Same with textures - they are being mapped over objects. if the same pattern is repeated you're seeing a render not a natural image. And you find repetitions by simplifying the image i.e. contrast which does the number trick explained above.
Remember - you are very rarely looking at an image. Your brain is matching the image on the retina with something that it already has in its memory. Even complex images work like that. The whole thing about learning drawing, painting or good photography is to unlearn that habit and just see shapes, colours, contrasts etc.
Last edited:
You open it on the 24th too?
- Status