Fairly sure that’s a PVM in the photo, which has no comb filter and is thus displaying the raw signal- that should be as blurry as it gets.
We need someone with a PVM to post a reference photo showing composite video.
Pretty sure the blur on CRTs was almost entirely attributable to beam misalignment on lower quality CRTs (which also usually only had composite video).
I know my NES composite output on my Sanyo CRT provides a sharp picture - more sharp than the composite shader presets I’ve tried but it’s a newer CRT from 2003. Probably has great comb filters, etc. I have a Trinitron curved CRT in my basement from 1991 though. I should see how my NES looks on that.
probably still sharper than what the shaders look like, I’m guessing. I think the shaders are emulating the signal as if no comb filter is applied… so like a TV from the 1970s, I guess.
It’s a good question how the dithering patterns are blended with a sharper appearance on a real CRT. Current ntsc adaptive shaders are set up to blend even more demanding dithering like the Sonic waterfall.
I guess it’s important to get info on this matter.
Yes we really need more empirical info on this, as in reference photos of CRTs displaying composite video.
From this post, this is composite output from a Wii on a non-ancient CRT:
And this is RGB output on the same CRT:
The composite output on that CRT looks very similar to cranking up the NTSC resolution scaling in guest-advanced with blend mode 2. It would appear more modern CRTs don’t actually blend those colors the way people expect. They’re sharper, but the waterfall clearly isn’t blended to the extent most people expect.
Edit
I’m not sure if the CRT is PAL or NTSC 
So this might be comparing apples to oranges. I just asked the poster to clarify.
According to the post by Nerdlypleasures I was linking in the composite artifacts thread a while ago, “raw” (Notch filter) output should be pretty much the standard case for 80’s 15 khz monitors, e.g the Commodore ones.
It’s my understanding that Sony PVMs also have no comb filter.
Pretty sure Hunterk posted a picture once when you two were discussing the same issue?
Anyway, here’s also a gallery with Nes composite on a Commodore 1701 https://imgur.com/gallery/xF2jP
And here is somebody complaining about Sega composite comparing NES composite presumably from a a TV:
We need better quality photos showing what’s going on at the subpixel level, I think.
What I’ve noticed when using my shader presets paired with my custom Blargg NTSC Filter presets is that you can reduce the blurring while still blending the Sonic The Hedgehog Waterfalls by cranking up the Horizontal and Subtractive Sharpness and also GTU resolution to the max or disabling GTU.
This results in an extremely harsh image that might be unfit for playing as is but perfect to now be an input for the various NTSC Filters.
Be sure to click on the images then Load High Resolution and zoom in until they look right.
I think your suggestion nearly contains the answer. I sometimes drop by my city’s only arcade bar to remind myself what crts are supposed to look like (albiet without composite video) and I’ve noticed how sharp centipede can be.
Games as ancient as centipede tend to have a few “pure” colors that use only one sub pixel. This creates a gap bewteen pixels because the other two sub pixels aren’t active (in addition to forming noticeable lines because centipede is TATE, but I digress). The physical space between pixels seems help create a shaper image.
Or to put this in short: THE POWER OF NATIVE RESOLUTION!
Edit: In retrospect the “native” in those words is kinda stupid considering what composit video does to a signal.
This image depicts a display with a comb filter. As do these two.
With a simple notch filter, uniformly-colored horizontal lines are perfect but vertical lines show artifacts. 2D comb filters produce perfect vertical lines but artifacts can appear in horizontal lines. The first image I linked probably shows a display with a fairly good adaptive 3-line comb filter: the strongest artifacts only occur when the color of one scanline is different from the scanlines above and below. My guess is that the display in the other images has a two-line comb filter, since the artifacts are not symmetric above and below a color transition.
I can’t figure out the mid-beam swelling that happens when using the NTSC shader. It seems like it’s sampling something in the vertical direction? I think it can really mess up some small details, like the small letters on this screen. All of the white lines that make up the letters should be a consistent width; the width of the picture line on a CRT is determined solely by the brightness of the pixel and nothing gets sampled or average in the vertical direction.
A CRT displaying a composite/NTSC signal:
The two pictures definitely have different scanline and horizontal filtering “setups”. 
Nice beam dynamics are resulting from blend mode 2.0, which consequences a different scanline feel, mostly due the original horizontal interpolation from the ntsc shaders.
What should I change to eliminate the beam swelling and achieve a consistent beam width per color? Guest-advanced doesn’t have this issue, btw.
shaders = "17"
shader0 = "shaders_slang/misc/grade.slang"
filter_linear0 = "false"
wrap_mode0 = "clamp_to_border"
mipmap_input0 = "false"
alias0 = ""
float_framebuffer0 = "false"
srgb_framebuffer0 = "false"
scale_type_x0 = "source"
scale_x0 = "1.000000"
scale_type_y0 = "source"
scale_y0 = "1.000000"
shader1 = "shaders_slang/crt/shaders/guest/advanced/stock.slang"
filter_linear1 = "false"
wrap_mode1 = "clamp_to_border"
mipmap_input1 = "false"
alias1 = "StockPass"
float_framebuffer1 = "false"
srgb_framebuffer1 = "false"
scale_type_x1 = "source"
scale_x1 = "1.000000"
scale_type_y1 = "source"
scale_y1 = "1.000000"
shader2 = "shaders_slang/crt/shaders/guest/advanced/afterglow0.slang"
filter_linear2 = "true"
wrap_mode2 = "clamp_to_border"
mipmap_input2 = "false"
alias2 = "AfterglowPass"
float_framebuffer2 = "false"
srgb_framebuffer2 = "false"
scale_type_x2 = "source"
scale_x2 = "1.000000"
scale_type_y2 = "source"
scale_y2 = "1.000000"
shader3 = "shaders_slang/crt/shaders/guest/advanced/pre-shaders-afterglow.slang"
filter_linear3 = "true"
wrap_mode3 = "clamp_to_border"
mipmap_input3 = "false"
alias3 = "PrePass0"
float_framebuffer3 = "false"
srgb_framebuffer3 = "false"
scale_type_x3 = "source"
scale_x3 = "1.000000"
scale_type_y3 = "source"
scale_y3 = "1.000000"
shader4 = "shaders_slang/crt/shaders/guest/advanced/ntsc/ntsc-pass1.slang"
filter_linear4 = "false"
wrap_mode4 = "clamp_to_border"
frame_count_mod4 = "2"
mipmap_input4 = "false"
alias4 = ""
float_framebuffer4 = "true"
srgb_framebuffer4 = "false"
scale_type_x4 = "source"
scale_x4 = "4.000000"
scale_type_y4 = "source"
scale_y4 = "1.000000"
shader5 = "shaders_slang/crt/shaders/guest/advanced/ntsc/ntsc-pass2.slang"
filter_linear5 = "true"
wrap_mode5 = "clamp_to_border"
mipmap_input5 = "false"
alias5 = ""
float_framebuffer5 = "false"
srgb_framebuffer5 = "false"
scale_type_x5 = "source"
scale_x5 = "0.500000"
scale_type_y5 = "source"
scale_y5 = "1.000000"
shader6 = "shaders_slang/crt/shaders/guest/advanced/custom-fast-sharpen.slang"
filter_linear6 = "true"
wrap_mode6 = "clamp_to_border"
mipmap_input6 = "false"
alias6 = "NtscPass"
float_framebuffer6 = "false"
srgb_framebuffer6 = "false"
scale_type_x6 = "source"
scale_x6 = "1.000000"
scale_type_y6 = "source"
scale_y6 = "1.000000"
shader7 = "shaders_slang/crt/shaders/guest/advanced/convert-ntsc.slang"
filter_linear7 = "true"
wrap_mode7 = "clamp_to_border"
mipmap_input7 = "false"
alias7 = ""
float_framebuffer7 = "false"
srgb_framebuffer7 = "false"
scale_type_x7 = "source"
scale_x7 = "0.500000"
scale_type_y7 = "source"
scale_y7 = "1.000000"
shader8 = "shaders_slang/crt/shaders/guest/advanced/stock.slang"
filter_linear8 = "true"
wrap_mode8 = "clamp_to_border"
mipmap_input8 = "true"
alias8 = "PrePass"
float_framebuffer8 = "false"
srgb_framebuffer8 = "false"
scale_type_x8 = "source"
scale_x8 = "1.000000"
scale_type_y8 = "source"
scale_y8 = "1.000000"
shader9 = "shaders_slang/crt/shaders/guest/advanced/avg-lum-ntsc.slang"
filter_linear9 = "true"
wrap_mode9 = "clamp_to_border"
mipmap_input9 = "true"
alias9 = "AvgLumPass"
float_framebuffer9 = "false"
srgb_framebuffer9 = "false"
scale_type_x9 = "source"
scale_x9 = "1.000000"
scale_type_y9 = "source"
scale_y9 = "1.000000"
shader10 = "shaders_slang/crt/shaders/guest/advanced/linearize-ntsc.slang"
filter_linear10 = "true"
wrap_mode10 = "clamp_to_border"
mipmap_input10 = "false"
alias10 = "LinearizePass"
float_framebuffer10 = "true"
srgb_framebuffer10 = "false"
scale_type_x10 = "source"
scale_x10 = "1.000000"
scale_type_y10 = "source"
scale_y10 = "1.000000"
shader11 = "shaders_slang/crt/shaders/guest/advanced/gaussian_horizontal.slang"
filter_linear11 = "true"
wrap_mode11 = "clamp_to_border"
mipmap_input11 = "false"
alias11 = ""
float_framebuffer11 = "true"
srgb_framebuffer11 = "false"
scale_type_x11 = "absolute"
scale_x11 = "640"
scale_type_y11 = "source"
scale_y11 = "1.000000"
shader12 = "shaders_slang/crt/shaders/guest/advanced/gaussian_vertical.slang"
filter_linear12 = "true"
wrap_mode12 = "clamp_to_border"
mipmap_input12 = "false"
alias12 = "GlowPass"
float_framebuffer12 = "true"
srgb_framebuffer12 = "false"
scale_type_x12 = "absolute"
scale_x12 = "640"
scale_type_y12 = "absolute"
scale_y12 = "480"
shader13 = "shaders_slang/crt/shaders/guest/advanced/bloom_horizontal.slang"
filter_linear13 = "true"
wrap_mode13 = "clamp_to_border"
mipmap_input13 = "false"
alias13 = ""
float_framebuffer13 = "true"
srgb_framebuffer13 = "false"
scale_type_x13 = "absolute"
scale_x13 = "640"
scale_type_y13 = "absolute"
scale_y13 = "480"
shader14 = "shaders_slang/crt/shaders/guest/advanced/bloom_vertical.slang"
filter_linear14 = "true"
wrap_mode14 = "clamp_to_border"
mipmap_input14 = "false"
alias14 = "BloomPass"
float_framebuffer14 = "true"
srgb_framebuffer14 = "false"
scale_type_x14 = "absolute"
scale_x14 = "640"
scale_type_y14 = "absolute"
scale_y14 = "480"
shader15 = "shaders_slang/crt/shaders/guest/advanced/crt-guest-advanced-ntsc.slang"
filter_linear15 = "true"
wrap_mode15 = "clamp_to_border"
mipmap_input15 = "false"
alias15 = ""
float_framebuffer15 = "true"
srgb_framebuffer15 = "false"
scale_type_x15 = "viewport"
scale_x15 = "1.000000"
scale_type_y15 = "viewport"
scale_y15 = "1.000000"
shader16 = "shaders_slang/crt/shaders/guest/advanced/deconvergence-ntsc.slang"
filter_linear16 = "true"
wrap_mode16 = "clamp_to_border"
mipmap_input16 = "false"
alias16 = ""
float_framebuffer16 = "false"
srgb_framebuffer16 = "false"
scale_type_x16 = "viewport"
scale_x16 = "1.000000"
scale_type_y16 = "viewport"
scale_y16 = "1.000000"
g_crtgamut = "0.000000"
g_vignette = "0.000000"
pre_bb = "0.750000"
quality = "-1.000000"
CONTR = "0.050000"
blendMode = "2.000000"
GAMMA_INPUT = "1.000000"
glow = "0.000000"
brightboost = "1.699999"
brightboost1 = "1.699999"
gsl = "2.000000"
beam_min = "1.000000"
beam_max = "1.200000"
beam_size = "0.000000"
vertmask = "0.500000"
scans = "1.999999"
scan_falloff = "2.000000"
spike = "0.000000"
h_sharp = "5.199976"
s_sharp = "0.500000"
maskstr = "1.000000"
mcut = "1.000000"
mclip = "0.000000"
gamma_out = "1.000000"
post_br = "1.200000"
textures = "SamplerLUT1;SamplerLUT2;SamplerLUT3;SamplerLUT4"
SamplerLUT1 = "shaders_slang/crt/shaders/guest/advanced/lut/trinitron-lut.png"
SamplerLUT1_linear = "true"
SamplerLUT1_wrap_mode = "clamp_to_border"
SamplerLUT1_mipmap = "false"
SamplerLUT2 = "shaders_slang/crt/shaders/guest/advanced/lut/inv-trinitron-lut.png"
SamplerLUT2_linear = "true"
SamplerLUT2_wrap_mode = "clamp_to_border"
SamplerLUT2_mipmap = "false"
SamplerLUT3 = "shaders_slang/crt/shaders/guest/advanced/lut/nec-lut.png"
SamplerLUT3_linear = "true"
SamplerLUT3_wrap_mode = "clamp_to_border"
SamplerLUT3_mipmap = "false"
SamplerLUT4 = "shaders_slang/crt/shaders/guest/advanced/lut/ntsc-lut.png"
SamplerLUT4_linear = "true"
SamplerLUT4_wrap_mode = "clamp_to_border"
SamplerLUT4_mipmap = "false"Tough one, if you want to use blend mode 2.0. I would suggest a compromise by maybe not going for a complete de-dithering effect by using different blend modes or/and increasing ‘ntsc resolution’.
Original ntsc shaders do a very wide filtering with gamma 1.0, which contributes to the effect. Any other gamma (2.4…) like in the regular version would need to be encoded specially, most likely to produce other unwanted looks/feels.
Looking at the parameter values, input/output gamma of 1.0 and scanline falloff also contribute to the effect…
