New CRT shader from Guest + CRT Guest Advanced updates

[guest.r]

Slotmask horizontal width shouldn’t be an issue for banding, it’s more or less to be aligned with crt mask width.

Actual vertical spacing comes in steps of two (must be an even number). As it can be seen the most problematic part is the interaction with scanlines. Quick cure for this is to use a smaller height or less pronounced scanlines.

Integer scaling can help, as it comes with two ‘flavors’ with the shaders, one bringing odd and the other even vertical original resolution scaling.

Ofc. does a display with higher resolution help a lot.

[guest.r]

New Release Version (2022-05-24-r1):

Notable changes:

• post brightness parameter step is now 0.01
• gamma bug with scanline gamma and interlace modes fixed (ntsc, hd versions)
• small cleanup
• ntsc shaders register usage reduced (unused parameter cleanup)
• nicer artifacting with ntsc-sharpness
• ntsc-fast shaders filtering reworked, not compatible with older version preset/parameter wise
• ntsc-fast version blending modes reworked (no longer needed, made obsolete by new features)
• edit: ntsc-fast preset also changed

Download link:

https://mega.nz/file/Ulo3WIpT#w17zeDk4yOBGrQ__nXp8SbYpaUS1pL3LarJwtiiRzb0

[RealNC]

Just noticed that ntsc-fast produces a ghost image when using horizontal overscan. Vertical overscan doesn’t do that.

Examples:

(Visible at the sides of the image.)

(Visible in the text in the upper-left corner. There’s ghost images of the yellow and white letters.)

[guest.r]

@RealNC

Just noticed that ntsc-fast produces a ghost image when using horizontal overscan. Vertical overscan doesn’t do that.

Thanks for the bug find. I think i’ll get this one soon… It should have something to do that source size is doubled horizontally with the new version.

[Nesguy]

Oh boy, new toys!

I know what I’ll be doing instead of working today…

[guest.r]

New Release Version (2022-05-24-r2):

Notable changes:

• overscan related bugs fixed

Download link:

https://mega.nz/file/R5h01LSS#rOvIS-zKI68VlrNRDZ5nEe3XcuSz6taP6SwQ_Vkw7Xs

[Red-gehog]

I thought the idea of ​​implementing gdapt in the shader was brilliant! But I don’t think it mixes dither as perfectly as composite video does. It’s sad for me that sharpness undoes the pseudo transparency in Sonic and Shinobi 3 waterfalls. I ended up going back to version 2022-05-13-r1, I thought it was perfect for me.

Well I still think your shader is amazing! Sorry if that sounded arrogant, I’m just a layman.

2022-05-13-r1

Sonic-The-Hedgehog-Japan-Korea-220525-004624

[guest.r]

@Red-gehog

I ended up going back to version 2022-05-13-r1, I thought it was perfect for me.

Latest ntsc version doesn’t doesn’t differ from 2022-05-13-r1 by default settings or if you don’t want to (it has only better sharpness controls), plus it has some bugs fixed (GDAPT is disabled by default settings…). I’m mentioning this so you can use the latest version without second thoughts… Anyway, thanks for trying things out.

[c9f5fdda06]

Maybe it’s your skill or maybe it’s my cognitive bias but your latest NTSC presets look つごい There are a lot of noticeable improvements in sharpness. My 90’s “kitchen TV” Sony Trinitron (13") always looked sharp despite only using composite.

[Red-gehog]

Sharpening really works well! But it undoes the dither blend.

One thing I’ve noticed is that this sharpening feels less destructive. But the previous one (custom fast sharpen) it was possible to leave the amount of sharpening at zero and increase the strength, because the amount also undoes the dither blend.

I combined that with subtractive sharpening and got maximum sharpening without undoing the dither. Well maybe I may be expressing myself wrong, so I’ll leave my parameters here, because I may just not be able to adjust it properly, so I apologize.

2022-05-13-r1

quality = “0.000000”
ntsc_phase = “2.000000”
SHARPEN = “4.000000”
CONTR = “0.000000”
blendMode = “2.000000”
h_sharp = “1.000001”
s_sharp = “1.500000”

2022-05-24-r2

quality = “0.000000”
ntsc_phase = “2.000000”
ntsc_sharp = “4.000000”
blendMode = “2.000000”
s_sharp = “1.500000”

comparison

[Red-gehog]

Well I found out that I can use Raspberry with composite video, I’m looking for a Trinitron crt, I know that probably the composite of the Pi is different from the Mega Drive Genesis, but I want to see if I can sharpen and dither blend with composite video. I don’t know if photos could really help…

[guest.r]

@Red-gehog

Sharpening really works well! But it undoes the dither blend.

I think i’m on something here:

It’s not perfect but looks quite nice to me.

[Red-gehog]

Look great! Could you share the parameters please?

[rafan]

@guest.r @Nesguy

In case you might find this interesting, with regards to colour matching Sony wrote this very readable 7-page whitepaper on

Colour Matching Between OLED and CRT

It says that an identical colourimetric instrument measurement of colours on CRT and OLED (i.e. theoretically the colour perception should be identical for both devices) still yield a different human perception of the colours on both devices. This because the spectral distribution of the light sources in both devices differ.

Sony argues that for the colours to more accurately match perceptually, an adjustment has to be made on the OLED to the 1931 CIE coordinates of the D65 whitepoint, the so called “Judd-Vos” refinements to the 1931 CIE model.

Sony has implemented this adjustment as a single offset on their pro OLED BVM/PVM monitors (see page 7):

This offset based on the “Judd modified correction colour matching function” will be applied to the following models effective from 2012 autumn shipping. <…>

Offset Values (from reference white point): (Δx, Δy) = (-0.006, -0.011) (x, y) = (0.3067, 0.3180) for D65

From some further reading I understand the Judd-Vos refinement offset is different per display technology (LED, OLED, Plasma) and varies also a bit per model, such that the adjustment is not a specific single value but yields a range. See the document here for information on the adjustment range per technology, mostly page 8-10:

Towards Standardizing a Reference White Chromaticity for High Definition Television

I’m not sure if this “Judd-Vos” adjustment has become industry practice by now for imlementing in hardware, but with the focus of sites like RTINGS.COM etc on measuring exact D65 1931 CIE values, I suspect it is not industry practice to include the adjustment in hardware.

As such maybe it could be considered for the shader. That said, possibly the change is too marginal to really notice unless you’re in a Sony studio looking at their OLED BVM

[guest.r]

@Red-gehog

Look great! Could you share the parameters please?

It required a major rework with the shaders, but looks ready for release now. Handling the of the new feature should be quite simple…

@rafan

It says that an identical colourimetric instrument measurement of colours on CRT and OLED (i.e. theoretically the colour perception should be identical for both devices) still yield a different human perception of the colours on both devices. This because the spectral distribution of the light sources in both devices differ.

This makes perfect sense, i also guess this is a complicated topic which could also involve other flat panel display types, calibration is also a factor. My personal factor, for example, are my glasses with an anti-reflective coating, which also alter the perceptual white point.

@rafan

As such maybe it could be considered for the shader.

I guess a color profile is doable without much trouble.

[guest.r]

New Release Version (2022-05-26-r1):

Notable changes:

• adaptive sharpness feature added to ntsc options
• (works best with 2 phase coding, fringing is also supported)
• ntsc presets changed
• dithering shaders are replaced with the new adaptive sharpness feature

Download link:

https://mega.nz/file/BwIiWDYb#FdtQ_aB_vmj9ieJVwPt16eVego50YE1EGWZYHJi_6WM

[RealNC]

Btw, just in case people didn’t know. the -hd preset can also do some blending to take the unpleasant edge off sharp dithering patterns. Obviously it won’t do NTSC colors. But if that’s what you want (original colors rather than NTSC artifacting,) it seems it’s a viable alternative.

Set “internal resolution” to “1.40” for this to work. Example:

[guest.r]

@RealNC

Btw, just in case people didn’t know. the -hd preset can also do some blending to take the unpleasant edge off sharp dithering patterns.

Yes, it has a more advanced filter implemented. I added the same horizontal filter to the ntsc-fast version and it works great with the new sharpness features. It has to be set by a couple of parameters, but brings very nice results.

[Nesguy]

@guest.r

Did you do something to the color code with the latest NTSC shader? I think it looks better, but maybe it’s my imagination?

Now that I’m looking at blend mode 1, I’m not sure what the problem is? Looks like the pixels are blended correctly? My eyes are definitely playing tricks on me - I keep seeing gaps between the blended pixels that aren’t really there; zooming in confirms it: the pixels are merged. There may be a weird optical illusion occurring with the waterfall.

preview of my updated high tvl NTSC preset.

It’s quite close to this shot taken on a PVM, minus the rainbow effect. While the rainbows may be authentic to the hardware, I think it is completely reasonable to assume that this is an effect that was NOT intended by the artists.

CRT Pixels 📺 (CRTpixels)

[HyperspaceMadness]

Are there reasons we wouldn’t want this kind of filter with the regular NTSC version?