Anonymous at Sat, 25 Mar 2023 23:00:04 UTC No. 941043

>>941036
you need to add dirt lenses, noise and other imperfect shit from IRL cameras.

most real photos usually have some shit like chromatic aberration, dust simulation, AO dust, dirt lenses, dust, shit tier resolution, amateur light conditions, overexposure, flash shit, small motion (because the finger press the button to shoot), lens distortion, DOF blur, lens flare, etc.

Ironically to make a render more real you need to basically add all the things that real phographers try to avoid.

Since 3D is too perfect for the real world, you need to add a lot of shit to make it actually more shit than what is acceptable on professional photography.

Anonymous at Sun, 26 Mar 2023 00:36:42 UTC No. 941053

>>941043
>dirt lenses
What shitty photographer leaves dirt on their lens?

Anonymous at Sun, 26 Mar 2023 02:01:40 UTC No. 941063

>>941053
your brain is used to seeing all the shitty photos your family takes.

the point is being subtle, you almost don't see it but where you can feel the diference.

Anonymous at Sun, 26 Mar 2023 02:05:02 UTC No. 941064

>>941053
also, your phone camera and IRL camera always will get some small dust particles or almost invisible dirt from your fingerprints.

Also, use some low fog simulation or volumetric fog to imitate air particles.

basically your renders are like filmed in a vacuum, so you need to like add shit like aereal perspective to the distance.

And remember to switch to filmic color space and try the most HDRI setting the imitatates the setting of an IRL camera.

And keep everything using the same focal lenses of an IRL camera.

would be nice to learn the diferences between 24mm, 50mm, 80mm and 150mm focal angles.

Anonymous at Sun, 26 Mar 2023 02:09:26 UTC No. 941065

>>941064
my production uses ACES

Anonymous at Sun, 26 Mar 2023 02:33:44 UTC No. 941068

>>941036
Me.
Though I've yet to find that "perfect" noise that looks like proper photography. It pisses me off to no end. Same for animated noise.
It's not enough for me to just overlay it on top, it has to be perfect. Noise isn't just overlayed on top of an image.
I still haven't found a good solution.

I'm not going balls to the wall trying to smear vaseline over the 3d camera or replicate everything that you'd avoid in photography. I come from a partly photographer background, and noise/grain is something that you kind of lean into and embrace in certain disciplines of photography.
Working in 3d really makes that hard to get believable noise and grain.
Both because I haven't found a pattern that looks perfect, and I haven't found a way of blending it that looks perfect. Overlay and multiply only affect the light parts of the scenes, while adding only affects the dark parts. There's no interaction.

Anonymous at Sun, 26 Mar 2023 09:26:16 UTC No. 941115

>>941068
>while adding only affects the dark parts
That's because you have to scale the result.
If an image is composed of decimal values between 0 and 1, the noise you're adding is also decimal values between 0 and 1, and the final JPG/PNG can only store decimal values between 0 and 1, then "ADD"ing noise pushes some of the image's pixels to be above 1. On export, those pixels get clipped to 1 and the visual result is bright spots that look like they didn't have any noise added to them.
Using multiply causes values near 0 to shrink to a value in between the nearest two quantization steps. This causes them to get rounded to 0 or their original value.
Scaling would involve multiplying, uniformly across all channels, the reciprocal of the theoretical highest pixel value (the highest value of your image + the highest value of your noise).

You could both multiply and add and then use the original image's mid-level value to mask which operation gets used (add for values below 0.5 and multiply for values above).

Another approach would be to both add and subtract noise to an image: to combine it with TPDF noise.
You would have some noise pattern with values spanning 0 to 1. You would multiply that image by 2 and subtract 1. Now it's values span -1 to 0 to 1. When you add this to your image, some pixels will be brightened (+1), remain unchanged (+0), and get darker (+(-1)). To change how much your noise shifts your image's pixels, you would multiply that -1/0/1 noise with some value between 0 and 1; multiplying by 0.123 would cause the noise to span -0.123 to 0 to +0.123.

As for patterns, I've found that a scaled-down image of perlin noise gives a better approximation of digital camera noise than what white noise would give.
For film noise/grain, you could always just use high-quality scans of film grain floating around the web.

Anonymous at Sun, 26 Mar 2023 10:27:48 UTC No. 941117

>>941036
https://soundcloud.com/xmfprox/soda-prod-teizeko

Anonymous at Mon, 27 Mar 2023 20:48:56 UTC No. 941304

>>941115
>If an image is composed of decimal values between 0 and 1, the noise you're adding is also decimal values between 0 and 1,
Pretty sure clamping the result or using exr mitigates that issue entirely, I think. Since you wouldn't be doing PP on jpg or PNG files. If you were though, in something like PS, it automatically clamps values as well based on the bit depth you pick. I'm sure stuff like AE and other video editors do as well. Generally the values clip automatically at 0-1 unless you're working in a format that goes beyond that.

Add affecting only bright spots is just how that blending mode works, and how you expect it to work. Since you're adding the bright part of the image, the bright parts added with the brights of another image add up and clamp to be white, and the dark parts get brighter because their values go up. Same with multiply but since you're 0-1, you'd be multiplying by decimals (the dark areas) and making the image darker since values would be lower.
At least that's how I understand it.

>You could both multiply and add.... Another approach would be to both add and subtract noise to an image
I do like these ideas though. I don't know why I hadn't thought of that. Thanks. Might end up with too much noise, but that should be easy to dial back I think.

>For film noise/grain, you could always just use high-quality scans of film grain floating around the web.
Yeah I guess that works too. For some reason I always figured in my head it would be hard to do since you'd have to take a video of a dark screen and then noise wouldn't show up.

Thanks for the tips bro.