A new little shader i did (glsl)

[DariusG]

Here is another composite secret on these 2 screenshots.

Raw rgb, dithering is pretty clear and obvious

A gray scale representation, look how dithering is almost gone invisible, how close dithered pixels are. Not by luck, that’s the artist’s brilliance there. This is our luma representation in a composite signal. Add some bluriness to this due to luma filters, add the colors of chroma on top, that is more like drawing with a crayon or something due to low bandwidth and you got dithering blend.

In conclusion, when Luma dithered pixels difference is large it won’t be able to blend. When in S-Video the signal evades some Composite Luma filters as it is clean already so it stays sharper, not blending that easily as in composite.

[DariusG]

raw

Real early Genesis model 1 Composite screen capture

crt-consumer-1w-ntsc-XL after reducing Luma bandwidth a bit (and increase comb filter)

[DariusG]

@ROBMARK85

Some small changes and added reflection from crt-cyclon

https://github.com/metallic77/shaders_glsl-slang/blob/main/1084s.zip

[ROBMARK85]

Even better with the reflection - thank you so much.

[Cyber]

@DariusG

added reflection from crt-cyclon

Will there ever be a fully modular reflection shader that can easily be appended / prepended to any shader / preset?

[DariusG]

@Cyber

Will there ever be a fully modular reflection shader that can easily be appended / prepended to any shader / preset?

Probably hard to implement without messing scanlines etc.

@ROBMARK85

Even better with the reflection - thank you so much.

Works better on integer-overscale, “mask” aligns better and creates a more convincing effect.

[ROBMARK85]

Thanks for the clarification and for your work - in fact using integer scale max is even better - great work and thanks

[DariusG]

A note about masks and why they are mostly inaccurate on most shaders at 1080p (the reason they feel “weird”, you can feel something is wrong).

The gap between the triad of pixels is around the same size of one R/G/B strip, in reality it’s around 20% of the total pixel size. Long story short, in order to be accurate the gap should be minimum 1 pixel and 3 pixels should be used for RGB stripes on every screen, that makes 270 TVL accurate pixels possible on a 1080p, 360 TVL on 1440p and so on. Every other way e.g. CGWG mask etc will create a mosquito mesh pattern as the gap will be same size with all 3 RGB stripes.

270 TVL but accurate on 1080p, around what a cheap low end 14" crt would do.

while 1440p will start looking more like a 20" crt at 360 TVL.

A 4K will do 2160:4 = 540, around a 14" PVM or 1084S but accurate.

So when using another mask like CGWG on 1080p we’re only trying to force the monitor to do something that isn’t capable of doing. The “pixels” are not accurate in any way and will feel weird. Should always be Red-Green-Blue-Black on every screen. Or the other way on BGR panels.

[Cyber]

@DariusG

Or the other way on BGR panels.

Or RBG on Pre 2026 LG WOLED Panels.

Sony Megatron Colour Video Monitor

The subpixels only matter as long as they’re visible to the eye, which is a function of viewing distance, pixel density and visual acuity (within reason; videophilia/audiophilia “golden” body parts notwithstanding). Beyond that threshold, we just draw the mask in whole pixels and move on. Based on Trogglemonkey’s calculations, we should be able to do that at around 6K res for non-weird CRTs (i.e., very high pitch). Based on Apple’s “retina” display designations, it looks like a PPI between 200 …

…and tentatively BRG on Post 2026 LG WOLED Panels.

[DariusG]

Well yes, whatever pattern is suitable for each monitor. But should be 3 pixels for RGB and 1 for black.

[DariusG]

If we research a bit, a Commodore 1702 has 0.64mm dot pitch on 14"

https://crtdatabase.com/crts/commodore/commodore-1702a

A 1080p at 15" has 0.17mm. On the above triad mask we saw that around 1/4 of the actual CRT pixel is occupied by the gap. That means 0.64 : 4 = around 0.16mm. Almost perfect for a 1080p, 1 pixel gap and 3 R/G/B pixels will do proper emulation of that monitor. Total 0.68mm of 1080p will be used for one emulated 1702 pixel. Anything less than 0.64mm dot pitch can’t be accurately emulated by a 1080p monitor.

(Not to mention it isn’t bright enough to do all this properly).

[kokoko3k]

@DariusG

The gap between the triad of pixels is around the same size of one R/G/B strip, in reality it’s around 20% of the total pixel size.

It is true that the gap between triads is low, but there is a gap after each component. From the pic it seems 65% phosphors, 35% gaps (which is indeed about 20*3)

@DariusG

Should always be Red-Green-Blue-Black on every screen.

Speaking of accuracy, if I’m right, doing rgbx on an underlying rgb grid gives:

Rxx|xGx|xxB|xxx -> rxxxgxxxbxxx, that’s a 3:1 ratio in favour of gaps!

I think the most “accurate” thing on common RGB/1080p panels is still green/magenta, which at least gives a fair 50:50:

indeed, xGx|RxB -> xgxrxb, gives perfectly spaced pixels but an high TVL of 1440/2=720

Another option would be rbg on RGB matrix: rxx-xxb-xgx, which could be better than rgb on rgb, because when the latter repeats, it couples the last b with the first r, giving weird purple patterns: rxxxgxxxbrxxxgxxxb

…but the eye always has the last word, with or without prescription glasses, in absence of which a special preset may be needed

[DariusG]

The problem with cgwg mask on a 1080p is that it will be 0.17mm for “pixel” and 0.17mm gap, no any real monitor ever did that. Lowering the in between rgb black gaps strength on r-g-b-black will be somewhat more convincing and true actual crt pixel size. Another alternative to save some brightness is to use just a black bar for the gap, cheap but still convincing. Irrelevant but i just recently discovered that

res = res + res * sin(vTexCoord.x * Outputsize.x * PI);

will hold brightness almost 100% even on full strength, as it will add 1.0 on position 0, and subtract 1.0 on position 1 (due to sine nature).

[kokoko3k]

@DariusG

0.17mm for “pixel” and 0.17mm gap,

I cannot figure out the 0.17mm dimension by gut, but I am assuming that 0.17mm “pixel” represent the single component and not the whole triad, right?

Green mageta result in: 0.17_,0.17G,0.17_,0.17R,0.17_,0.17B which is the closest thing to the real rgb layout of a crt, at least for the spacing, why not?

[DariusG]

A 1080p monitor at 15" has around 0.17mm dot pitch. That means one emulated pixel will be 0.17mm size and same will be for the gap. While e.g. a 1702 monitor has 0.64mm at almost same inches, around 4 times wider in the eye. A 1084S is 0.42mm. a cgwg mask will be closer to 1084s size, only problem is that the 1084s gap will be like 0.08

[DariusG]

Just a heads up, i updated “tiny-ntsc” glsl for the moment, to a better, more accurate/realistic ntsc simulator but still quite fast. By using a 4x horizontal scale initially we can easily manage to merge dithering and still remain sharp. This is what you would get on a really good composite like the Amiga or a modern raspberry pi, very sharp and still merging dithering. Not trying to mimic nes etc, only the default ntsc specs, which are pretty close to what i get on my Amiga/rpi.

[DariusG]

It’s totally possible to use scanlines and mask and keep almost full brightness even on a normal monitor by exploiting sine wave -1,+1 nature (one line adds 1, next subtract 1, so result is even) . Of course not full strength on bright areas to avoid clipping. Some vivid crt-like colors too

[saeedalshaikhpygas]

That does look bright, is this preset out yet or is it still wip?

[DariusG]

I injected the formula to crt-consumer-arcade (not yet approved), i usually create a ton of new shaders experimenting and pass what is worthy to existing ones.

crt-consumer-arcade, use scanlines/mask that won't decrease screen brightness

There is a gamma 2.2 on this shader too, so it deliberately darkens dark colors a bit.

[DariusG]

Difference in brightness, raw, crt-cgwg-fast and crt-consumer-arcade. Losing most dark details in crt-cgwg and similar shaders, while using this “trick” will be almost 1:1 brightness with raw.

Notice crt-cgwg loses a ton of brightness while it uses a mask that’s barely visible when crt-consumer-arcade uses a mask that is almost full strength. Because crt-cgwg uses an effect that is only subtracting while on the other hand we use “subtract-evenly add” and so on.