Trying to upscale textures to 4k results in black textures

[wyman_hd 232]

So i was trying to upgrade the standard definition textures of the Coolsky DC-9 to 4k using an AI upscaler. And it works fine, the textures themselves look beautiful. However, the textures refuse to appear in P3D. The native resolution of the textures is 1024 x 1024 and i upscaled it to 4096 x 4096. I exported it as a .jpg from the original texture using Photoshop, upscale the textures, (because the program i use cannot upscale .DDS files) then put it into photoshop and then export it back as a .DDS. And it works fine, but it will appear black in P3D. I have a 1080p monitor, but I'm improving the texture resolution, so I won't necessarily need a 4K monitor (at least i don't think). The only way to fix this is to change the resolution BACK to 1024 x 1024 then it will appear. I don't need to "downscale" the textures, i just have to change the texture size. I upscaled the normal maps and bump maps as well, but it still shows as black in P3D.

**EDIT**
Sorry, it has been fixed. It was at the wrong resolution, so i put it back to 4096 x 4096 and it works !

Edited September 11, 2019 by arwasairl

[Enko 46]

Try this

open the dds in Photoshop, do image > image size then 4096

[wyman_hd 232]

That will result in a blurry picture without upscaling. The whole point of upscaling is to improve the quality.

[cmpbellsjc 2,960]

6 hours ago, arwasairl said:

That will result in a blurry picture without upscaling. The whole point of upscaling is to improve the quality.

Upscaling a texture from 1024 to 4096 won’t make it more defined because your starting with a small texture sheet. You can get away with down scaling from 4096 to 1024 because your already starting with a higher definition texture sheet. Upscaling can’t magically add definition that wasn’t there before. That would be like trying to upscale a video that was shot in 720 and upscaling it to 1080, it will just end up looking blurry and pixelated.

You can also think of taking a photo that was printed as a 4x6 and blowing it up to the size of an 8x10. Sure the 8x10 will be larger but it won’t be as sharp as the original 4x6.

[virtuali 4,419]

5 hours ago, cmpbellsjc said:

You can also think of taking a photo that was printed as a 4x6 and blowing it up to the size of an 8x10. Sure the 8x10 will be larger but it won’t be as sharp as the original 4x6.

Yes, in general this is true but, there's a new generation of AI-based upscalers, that really do wonders when enlarging low res images, for example:

https://topazlabs.com/gigapixel-ai/

[wyman_hd 232]

9 hours ago, cmpbellsjc said:

Upscaling a texture from 1024 to 4096 won’t make it more defined because your starting with a small texture sheet. 

Here's a side-by-side comparison of a Delta texture for the DC-9 and using an AI upscaler to upscale it.

Sure, the new one is not perfect, but it makes the edges less blurry. 
I'm not trying to blow up a texture, rather using AI to try to approximate definitions, making the texture resolution appear larger.

Edited September 12, 2019 by arwasairl

[cmpbellsjc 2,960]

11 hours ago, virtuali said:

Yes, in general this is true but, there's a new generation of AI-based upscalers, that really do wonders when enlarging low res images, for example:

https://topazlabs.com/gigapixel-ai/

Interesting, I had no idea. I was wondering where the extra pixels would come from without redrawing a new texture sheet.

[virtuali 4,419]

8 hours ago, cmpbellsjc said:

Interesting, I had no idea. I was wondering where the extra pixels would come from without redrawing a new texture sheet.

From the "training" which was made over a very large set of photographs, with as many as possible real world examples, taken in the most varied situations.

I read the theory behind it, and It's really fascinating: the AI is not taking "pieces" of other images, but it has learned HOW an image lose its quality when it's downsized, or how and how much quality is lost with compression so, when upscaling, it tries to apply that knowledge gained through training, to the actual situation.

Here, a very popular open source project, which is very suitable for improving older games textures, based on the same principles:

https://www.theverge.com/2019/4/18/18311287/ai-upscaling-algorithms-video-games-mods-modding-esrgan-gigapixel

Edited September 13, 2019 by virtuali

[Pete Dowson 558]

On 9/12/2019 at 7:06 AM, cmpbellsjc said:

Upscaling can’t magically add definition that wasn’t there before. That would be like trying to upscale a video that was shot in 720 and upscaling it to 1080, it will just end up looking blurry and pixelated.

Odd, then, that my normal Blu-Ray HD films look abosolutely fabulous on my large screen using a 4K projector which upscales. It does a marvellous job of refining the detail. 

Even bog-standard non-HD DVD films and TV shows look good too, but of course not with as much sharp detail.

I think there are a number of very clever algorithms at work, and, in the video media world, there are several different technologies doing the same thing, some possibly better than others.

Of course, even the usually applied anti-aliassing methods are visually sharpening edges which otherwise become jagged due to the textures being displayed at larger than their pre-defined sizes, in terms of pixels.

Pete