Issues in converting 480i video to 1080i

[sdx12349]

Well... Some reason, when I convert 480i source to 1080i in Premiere Pro, this happens. (Click to see original image, because preview image is worse resolution than original, so you couldn't see same phenomenon in preview.)

This is originally 480i Cable TV TS Capture, but I resized to 1080i using Premiere Pro.

and most of TV Channel in South Korea also have this problem after late 2023~2024.

This is original 1080i TV Broadcasting.

but, if I convert it to 480i again, it does not seem to original quality.
I thougnt It would be better deinterlace filter, but when I used QTGMC, it seems just the same using Bob filter.
How can I convert it to original quality or similar?
I couldn't found the way, except keep resize untill it seems good and use QTGMC... but it isn't perfect way.
I want to get 480i video that looking just same as original 480i source, not the progressive video.
Could anyone tell me what's the problem, and how can I recover it to original quality?

[aramkolt]

Hmm. I haven’t really heard of many use cases for 1080i in modern times other than perhaps certain older TVs that can do 1080i and not 1080p or 720p. I would think the better path might be upscaling/deinterlacing to 1080p at 50/59.94fps (depending on if PAL or NTSC) before it even goes into Premiere using QTGMC.

[sdx12349]

Hmm. I haven’t really heard of many use cases for 1080i in modern times other than perhaps certain older TVs that can do 1080i and not 1080p or 720p. I would think the better path might be upscaling/deinterlacing to 1080p at 50/59.94fps (depending on if PAL or NTSC) before it even goes into Premiere using QTGMC.

Well, the reason I'm having trouble with this is because I can't manipulate it. South Korea's TV Channel is also have this issue when they broadcast old show/animation, and they don't convert 480i to perfectly in 1080i. Even if I could find the same scaling value, there's another problem: the video quality isn't the same or similar as the original.

I want to get nearly the same quality as the 480i source from a Korean TV channel's strangely converted 1080i video.

I attached korean TV channel's weird SD to HD converted broadcast feed source in first thread, and in this thread I attached the 480i video that I adjusted as much as possible with the Premiere Pro resize filter.

(sorry for the bad english)

[sdx12349]

I know after I resize video and use QTGMC, It's quality is much better then before. but it is not I really want. I want to get clean 480i source.

[jagabo]

Something like this 480i to 1080i?

[sdx12349]

Something like this 480i to 1080i?

well.. this thing is pretty great, but not so same what I want. It looks like applied some filter, so not suitable in my standard...
I think I chose wrong thread title, What I really want is "Convert 1080i video (that weirdly resized 480i video) to most likely to 480i source". not 480p, because I can get better 480p video by using QTGMC.
To better help, I attached original 480i source video, and video that weirdly converted original 480i video to 1080i by me.
It is not same as korean TV Channels do because I'm not employee of Broadcasting Company, but I think it is similar condition.
I know It's hard challenge, but I really want to resolve this problem strongly.

[poisondeathray]

To make Example_1080I_SD.mov similar to Example_480i_Origianl.mov, separate the fields, resample (downscale), the correct the colors for SD, weave

Something like this (you can correct for the borders, or use other resampling algorithms if you want it sharper/less sharp)

LSmashVideoSource("Example_1080I_SD.mov")
AssumeTFF().SeparateFields()
DebilinearResizeMT(720,240)
Colormatrix(mode="rec.709->rec.601", clamp=0)
Weave()

The other versions posted earlier are slightly different

[sdx12349]

To make Example_1080I_SD.mov similar to Example_480i_Origianl.mov, separate the fields, resample (downscale), the correct the colors for SD, weave

Something like this (you can correct for the borders, or use other resampling algorithms if you want it sharper/less sharp)

LSmashVideoSource("Example_1080I_SD.mov")
AssumeTFF().SeparateFields()
DebilinearResizeMT(720,240)
Colormatrix(mode="rec.709->rec.601", clamp=0)
Weave()

The other versions posted earlier are slightly different

Well.. It's Pretty not bad! but... It's not different what I can do in Premiere Pro.. I tried similar way in Premiere, and got similar or better result.
There's nothing another way better than this?

[poisondeathray]

Well.. It's Pretty not bad! but... It's not different what I can do in Premiere Pro.. I tried similar way in Premiere, and got similar or better result.
There's nothing another way better than this?

They are all bad to some extent, but there are literally hundreds of ways you can do it...

In what way do you want it "better" ? Try to do a better description

Another common way is to double rate deinterlace, downscale, reinterlace. That would be similar to PP, but you have higher quality choices for deinterlacers and scaling algorithms

Even without other operations, and you disregard lossy compression, upscaling then downscaling , is a lossy process unless you use nearest neighbor scaling on powers of 2. The "source" examples definitely did not, so you will never get similar to REAL original source on fine detail, such as small text. Perhaps you can train an AI to recreate those in the future, but there is no conventional way to fix those details

[poisondeathray]

If you look at the 1st post examples of the credits /text e.g. "2 - 투니버스_20250809_235546.ts" , both fields are missing data on static text. There is no conventional way to fix that properly. Maybe some anime folks trained some AI , I'm not up to date on that

If it were me, I would IVTC and ignore the text. That way you get the highest quality progressive content for the main parts, and you don't degrade everything by deinterlacing. This is anime - the original content is 24p , not interlaced. If you need sd interlaced format, you can downscale progressively after IVTC, and reapply the pulldown for 29.97i

Gamma aware downscaling will reduce the loss on thin white or bright text lines and characters. e.g. in avisynth a simple way is to use ResampleHQ , but you can linearize with AvsResize as well, apply scaling, then delinearlize . But that will not necessarily help thin dark text as well as bright details or lines

LWLibavVideoSource("2 - 투니버스_20250809_235546.ts")
TFM(pp=0)
TDecimate()
ResampleHQ(720,480)
colormatrix(mode="rec.709->rec.601", clamp=0) # if you want SD color


#If you need to apply pulldown for 29.97i
#ChangeFPS(60000,1001)
#AssumeTFF().SeparateFields().SelectEvery(4,0,3).W eave()

[sdx12349]

If you look at the 1st post examples of the credits /text e.g. "2 - 투니버스_20250809_235546.ts" , both fields are missing data on static text. There is no conventional way to fix that properly. Maybe some anime folks trained some AI , I'm not up to date on that

If it were me, I would IVTC and ignore the text. That way you get the highest quality progressive content for the main parts, and you don't degrade everything by deinterlacing. This is anime - the original content is 24p , not interlaced. If you need sd interlaced format, you can downscale progressively after IVTC, and reapply the pulldown for 29.97i

Gamma aware downscaling will reduce the loss on thin white or bright text lines and characters. e.g. in avisynth a simple way is to use ResampleHQ , but you can linearize with AvsResize as well, apply scaling, then delinearlize . But that will not necessarily help thin dark text as well as bright details or lines

LWLibavVideoSource("2 - 투니버스_20250809_235546.ts")
TFM(pp=0)
TDecimate()
ResampleHQ(720,480)
colormatrix(mode="rec.709->rec.601", clamp=0) # if you want SD color


#If you need to apply pulldown for 29.97i
#ChangeFPS(60000,1001)
#AssumeTFF().SeparateFields().SelectEvery(4,0,3).W eave()

well, I guess I found very good example!
I found 2008~2011 480i re-air video and 2025 1080i re-air video.
I guess It could help this challenge a lot.

"학교괴담OP.ts" and "학교괴담ED.ts" files are 2025 1080i re-air video,
and "학교괴담 오프닝.ts" and "학교괴담 엔딩.ts" files are 2008~2011 480i re-air video.
**This videos' first broadcast date is 2002.

[sdx12349]

and, If you convert attached 1080i videos to 480i without any adjust, It looks much worse than mine.
so I adjusted video again in Premiere Pro. and I attached Preset and result video, so check it if you want.
and can I use other better resizing filter than now?
I think Premiere Pro's resizing filter is the reason why video is so pixelated.

[sdx12349]

I just found out that I couldn't get similar result which I already got in Premiere Pro's resize filter.
I don't understand why only Premiere Pro get better result than others...
or TV Companies are using Premiere Pro to edit 480i video fit into 1080i, therefore only the Premiere Pro could turn back SD 1080i source similar to 480i original source?

[poisondeathray]

I think Premiere Pro's resizing filter is the reason why video is so pixelated.

Aliasing artifacts - It's partly because of the bad upscale and resizing the artifacts already present, partly because of PP deinterlacing causing new artifacts

I just found out that I couldn't get similar result which I already got in Premiere Pro's resize filter.
I don't understand why only Premiere Pro get better result than others...

The PP result has aliasing artifacts. Is that supposed to be "better" ?

or TV Companies are using Premiere Pro to edit 480i video fit into 1080i, therefore only the Premiere Pro could turn back SD 1080i source similar to 480i original source?

No , because it was a poor upscale to begin with and you just showed PP does a poor job on this source with more artifacts



How is this ?

[sdx12349]

I think Premiere Pro's resizing filter is the reason why video is so pixelated.

Aliasing artifacts - It's partly because of the bad upscale and resizing the artifacts already present, partly because of PP deinterlacing causing new artifacts

I just found out that I couldn't get similar result which I already got in Premiere Pro's resize filter.
I don't understand why only Premiere Pro get better result than others...

The PP result has aliasing artifacts. Is that supposed to be "better" ?

or TV Companies are using Premiere Pro to edit 480i video fit into 1080i, therefore only the Premiere Pro could turn back SD 1080i source similar to 480i original source?

No , because it was a poor upscale to begin with and you just showed PP does a poor job on this source with more artifacts



How is this ?

Yeah, I know that Premiere Pro result has Aliasing artifacts, but I couldn't get any better result than Premiere Pro when using AviSynth.
In AviSynth, all of resizing filter looks like attached image.(see in larger version)
I'm newbie in AviSynth, so I can't find any better way.

I'm curious what your method is and what kind of results you'd get if you applied it to an OP video rather than an ED video.

[poisondeathray]

Here is a non par corrected 1:1 screenshot of that frame for the premiere version you posted, and the avisynth filtered version . The difference in AR/borders is because PP has a specific interpretation for SAR for 4:3 DAR 720x480 , whereas the script I used just took the full frame and downscaled for the example. You can adjust it however you want.


[Attachment 89421 - Click to enlarge]


[Attachment 89422 - Click to enlarge]

I'm curious what your method is and what kind of results you'd get if you applied it to an OP video rather than an ED video.

You can try it

I used avisynth as I described above, but added some antialiasing. IVTC, descale with sharper parameters, and for antialiasing / fix artifacts - I used AnimeUndeint_Compact . This is just 1 way of doing it , there are dozens of variations for each filter. Many types of AA filters etc....

LWLibavVideoSource("학교괴담ED.ts")
TFM(pp=0)
Tdecimate()
UserDefined2ResizeMT(720,480, b=40,c=-20,s=2)
z_convertformat(pixel_type="RGBPS", use_props=0, colorspace_op="709:709:709:l=>rgb:709:709:f")
mlrt_ncnn(network_path="PATH\1x_AnimeUndeint_Compact_130k_net_g.onnx", builtin=false)
z_convertformat(pixel_type="YV12", use_props=0, colorspace_op="rgb:709:709:f=>170m:709:709:l") #601 colors

#Reinterlace - personally I wouldn't do this step and keep it progressive
ChangeFPS(60000,1001)
AssumeTFF().SeparateFields().SelectEvery(4,0,3).Weave() 

#Trim(79,2739) #trim to ED , cut commercials

If you want to keep commercials /live action, make sure to filter sections differently, because obviously they were processed differently


Taking a quick look at OP - you could probably filter it slightly differently , because there are additional problems, such as more compression artifacts, especially on scene changes. ( Also the other older 480i version had compression problems, probably none of the broadcast versions used proper master)

[sdx12349]

Here is a non par corrected 1:1 screenshot of that frame for the premiere version you posted, and the avisynth filtered version . The difference in AR/borders is because PP has a specific interpretation for SAR for 4:3 DAR 720x480 , whereas the script I used just took the full frame and downscaled for the example. You can adjust it however you want.


[Attachment 89421 - Click to enlarge]


[Attachment 89422 - Click to enlarge]

I'm curious what your method is and what kind of results you'd get if you applied it to an OP video rather than an ED video.

You can try it

I used avisynth as I described above, but added some antialiasing. IVTC, descale with sharper parameters, and for antialiasing / fix artifacts - I used AnimeUndeint_Compact . This is just 1 way of doing it , there are dozens of variations for each filter. Many types of AA filters etc....

LWLibavVideoSource("학교괴담ED.ts")
TFM(pp=0)
Tdecimate()
UserDefined2ResizeMT(720,480, b=40,c=-20,s=2)
z_convertformat(pixel_type="RGBPS", use_props=0, colorspace_op="709:709:709:l=>rgb:709:709:f")
mlrt_ncnn(network_path="PATH\1x_AnimeUndeint_Compact_130k_net_g.onnx", builtin=false)
z_convertformat(pixel_type="YV12", use_props=0, colorspace_op="rgb:709:709:f=>170m:709:709:l") #601 colors

#Reinterlace - personally I wouldn't do this step and keep it progressive
ChangeFPS(60000,1001)
AssumeTFF().SeparateFields().SelectEvery(4,0,3).Weave() 

#Trim(79,2739) #trim to ED , cut commercials

If you want to keep commercials /live action, make sure to filter sections differently, because obviously they were processed differently


Taking a quick look at OP - you could probably filter it slightly differently , because there are additional problems, such as more compression artifacts, especially on scene changes. ( Also the other older 480i version had compression problems, probably none of the broadcast versions used proper master)

Well I need to explain why I didn't use IVTC.
because Korean Anime Channel didn't edit video in 24p when they remove japanese and replace it to korean.
they used 60i in that, so if use IVTC, I think that looks jaggy or blurry. so this is not the correct answer.

sorry for my hard request, but It is not I want. I want to resize it without deinterlace or decimate.

[poisondeathray]

Well I need to explain why I didn't use IVTC.
because Korean Anime Channel didn't edit video in 24p when they remove japanese and replace it to korean.
they used 60i in that, so if use IVTC, I think that looks jaggy or blurry. so this is not the correct answer.

sorry for my hard request, but It is not I want. I want to resize it without deinterlace or decimate.

Do you think test1.mp4 looks jaggy or blurry ?

It is 29.97i . It's 23.96p in 29.97i . One difference might be some title fades and overlays could be at actual 59.94 field/s - ie true interlaced, but very few content is like this in those examples - You can separate the fields, or bob deinterlace and see. If at 59.94 there is a unique new frame every frame, then that would indicate true interlaced content - not very much of that in your example

Guess what? Premiere is deinterlacing when it resizes because video is flagged interlaced.

You can't (!!) progressively resize interlaced, or film content telecined to interlace without deinterlacing first, or separating the fields then resizing (as demonstrated earlier), or IVTCing film telecine . An interlace aware resize, either separates the field first (demonstrated in earlier post), or deinterlaces first. There is no other way

Deinterlacing is the reason it looks more jaggy and blurry. It adds to the jagginess and blurriness.

Good luck doing this without IVTC , deinterlace (premiere is deinterlacing) . The only option would be separating the fields demonstrated earlier. You choose

[sdx12349]

Well I need to explain why I didn't use IVTC.
because Korean Anime Channel didn't edit video in 24p when they remove japanese and replace it to korean.
they used 60i in that, so if use IVTC, I think that looks jaggy or blurry. so this is not the correct answer.

sorry for my hard request, but It is not I want. I want to resize it without deinterlace or decimate.

Do you think test1.mp4 looks jaggy or blurry ?

It is 29.97i . It's 23.96p in 29.97i . One difference might be some title fades and overlays could be at actual 59.94 field/s - ie true interlaced, but very few content is like this in those examples - You can separate the fields, or bob deinterlace and see. If at 59.94 there is a unique new frame every frame, then that would indicate true interlaced content - not very much of that in your example

Guess what? Premiere is deinterlacing when it resizes because video is flagged interlaced.

You can't (!!) progressively resize interlaced, or film content telecined to interlace without deinterlacing first, or separating the fields then resizing (as demonstrated earlier), or IVTCing film telecine . An interlace aware resize, either separates the field first (demonstrated in earlier post), or deinterlaces first. There is no other way

Deinterlacing is the reason it looks more jaggy and blurry. It adds to the jagginess and blurriness.

Good luck doing this without IVTC , deinterlace (premiere is deinterlacing) . The only option would be separating the fields demonstrated earlier. You choose

I thought that Premiere Pro didn't deinterlace in interlace sequence..
maybe the reason you thought like that was Premiere Pro's default viewport setting is "Display First Field".
if I choose "Display Both Field", It displays like weave, so Premiere Pro doesn't deinterlace when resizing interlaced video.
so Premiere Pro resizing is technically same what you've done, but diffrence is resizing algorithm i guess.
If I choose CPU acceleration mode in Premiere, resizing algorithm changes, and result is also different what I did in GPU acceleration mode.
It looks much same as your field separate and resize results.

and, the video that I attached in this discussion is not have 60i Korean edit thing, but I'll attach another example video. check it out.

[poisondeathray]

I thought that Premiere Pro didn't deinterlace in interlace sequence..
maybe the reason you thought like that was Premiere Pro's default viewport setting is "Display First Field".
if I choose "Display Both Field", It displays like weave, so Premiere Pro doesn't deinterlace when resizing interlaced video.
so Premiere Pro resizing is technically same what you've done, but diffrence is resizing algorithm i guess.
If I choose CPU acceleration mode in Premiere, resizing algorithm changes, and result is also different what I did in GPU acceleration mode.
It looks much same as your field separate and resize results.

Whenever premiere resizes interlaced material, it deinterlaces in the background. All video editors do this, not just premiere. Separating fields, and resize is a form of deinterlace - it's interpolating the missing scan lines (that's what resampling or resizing is doing). For progressive content (such as the actual background , not text), deinterlacing interpolates the other set of scanlines (odd or even) instead of combing both for a full frame, so in general you get ~50% resolution in general case compared to IVTC (this case is not general, it's a problematic upscale). But that's why you get additional jaggies and aliasing artifacts on both cases. The display setting does not affect the rendered final video

and, the video that I attached in this discussion is not have 60i Korean edit thing, but I'll attach another example video. check it out.

It does have "60i" content addition overlays but small amounts in fades. e.g. 학교괴담ED.ts if you double rate deinterlace to 59.94p, and examine around frame 3840-3850, the fade of the black text/yellow outline opacity changes every frame (originally a field now a frame)

So if that preservation of text fades is important to you , you are going to have to make some tradeoffs. In order to preserve 59.94 fields, you have to work in fields, or deinterlace, and that will degrade the progressive parts like the background, the anime art = aliasing, jaggies

[poisondeathray]

60i_example.ts is not interlaced content, at least the main content is not . It's still 23.976p content

For the text/graphics, I'm guessing they tried to overwrite the text with korean , and did a bad /sloppy job of the compositing? On top of the poor upscale / other problems

ie. the overlays don't motion track properly to cover up the original japanese in some parts . eg. If you double rate deinterlace, the building face sign, and later the box labels 60i_example.ts slip off the box on some parts . The very last part is ok (just the other problems with bad upscale)


If it was important, I'd fix the compositing , especially if was just a few scenes. The building could probably be ok just IVTCed, there is only a bit of slippage. The lateral box pan has problems and needs to be redone. Just redo it in AE. The left some japanese e.g. around 12 seconds you could fix that too and make it korean




60i_example2.ts is interlaced content. If the pan samples (smoothness) is imporant you can make it VFR if you wanted it mixed, or you can retime the pan at 24p , or redo the animation. Any type of deintearcing will cause aliasing on the text characters making them more difficult to read (or IVTC post processing like without TFM(pp=0), which will deinterlace)

60i_example3.ts is interlaced content. In order to keep the smoothness and 59.94 samples/sec you must work in fields or deinterlace - that means aliasing problems . Choose an algorithm that give you the least artifacts. There are "AI" ones too, such as MFDIN in vapoursynth. If you wanted to, you could make it 59.94p content mixed with 24p content VFR


You have to decide how much work you want to do, or if you want easy "1 script fits all" and make some tradeoffs, because you can't have good treatment on everything with out filtering sections differently

[sdx12349]

I found method that convert 1080i video almost similar to original 480i video!

LWLibavVideoSource("코난ED.ts")
interlaced_clip = last
top = interlaced_clip.SeparateFields().SelectEven()
bottom = interlaced_clip.SeparateFields().SelectOdd()
target_height_float = 242
target_height_int = 240
source_height_original = top.height
effective_source_height = target_height_int * source_height_original / target_height_float
crop_offset = source_height_original - effective_source_height
top_resized = top.ConvertToRGB32().UserDefined2ResizeMT(640, target_height_int, b=0, src_height=effective_source_height, src_top=crop_offset)
bottom_resized = bottom.ConvertToRGB32().UserDefined2ResizeMT(640, target_height_int, b=0, src_height=effective_source_height, src_top=crop_offset + 1)
Interleave(top_resized, bottom_resized)
Weave().AssumeTFF()

I used gemini because I'm newbie in avisynth, so script is looks dirty.. but I get most better result with seperate fields!
Important point was src_top=1 in bottom field resizing part.
but It has ghosting in some part. see in the image below.

[poisondeathray]

Improved, but still has some aliasing . You might be able to play with some parameters to get it closer

The artifact is generally referred "ringing" or "halos" from oversharpening, not "ghosting". Ghosting implies a temporal discrepancy, such as blend of 2 frames from 2 different times. You can try adjusting the resize parameters to a softer scaler (change the b, c, s values until you find something you like) . YOu have to make some tradeoffs between sharpness vs. ringing artifacts . I would avoid using "dehalo" filters, because they are quite damaging - try to fix it on the sharpening end, or there will be a lot of counterproductive back and forth

RGB32 is probably not required and more damaging than staying in YUV - if you need 1 pixel cropping on 4:2:0 material, YV24 (4:4:4) is probably better option

[sdx12349]

Yeah I removed RGB32, because it removes color details a lot.
and I modified b,c,s, but I couldn't get better result I guess.
I need to try more times?
I thought that I need to modity height float, not b,c,s values...

[poisondeathray]

Yeah I removed RGB32, because it removes color details a lot.
and I modified b,c,s, but I couldn't get better result I guess.
I need to try more times?
I thought that I need to modity height float, not b,c,s values...

The halos are 100% definitely from oversharpening, and for this case, you control that with the b,c,s values . You can try it, you will see it definitely reduce the halso. Preview script in avspmod or vdub2 (you don't have to waste time encoding until you are satisfied)

You have to decide if you are ok with the halos or want to reduce them. Personally I would reduce them because they are very distracting

[sdx12349]

Yeah I removed RGB32, because it removes color details a lot.
and I modified b,c,s, but I couldn't get better result I guess.
I need to try more times?
I thought that I need to modity height float, not b,c,s values...

The halos are 100% definitely from oversharpening, and for this case, you control that with the b,c,s values . You can try it, you will see it definitely reduce the halso. Preview script in avspmod or vdub2 (you don't have to waste time encoding until you are satisfied)

You have to decide if you are ok with the halos or want to reduce them. Personally I would reduce them because they are very distracting

than what value you recommend? I don't have idea about this...

[poisondeathray]

than what value you recommend? I don't have idea about this...

It's personal opinion - a tradoff between how sharp/blurry and how severe halo artifacts you want. For me, the halos are too strong right now, so different values than what you have now

Look at the documentation for ResampleMT for a short description and range of values. B and c go from -50 to 250

You can setup sliders in avspmod, and get basically instant feedback (like a dial or slider in premiere for filters) . Or just enter value and push f5 to refresh . Once you play with the numbers high, low , medium you will get a feel for what they do

[sdx12349]

than what value you recommend? I don't have idea about this...

It's personal opinion - a tradoff between how sharp/blurry and how severe halo artifacts you want. For me, the halos are too strong right now, so different values than what you have now

Look at the documentation for ResampleMT for a short description and range of values. B and c go from -50 to 250

You can setup sliders in avspmod, and get basically instant feedback (like a dial or slider in premiere for filters) . Or just enter value and push f5 to refresh . Once you play with the numbers high, low , medium you will get a feel for what they do

no, I just find out that height value was Important point.

LoadPlugin("LSMASHSource.dll")
LoadPlugin("ResampleMT.dll")
source = "코난ED.ts"
audio = LWLibavAudioSource(source)
LWLibavVideoSource(source)

interlaced_clip = last

top = interlaced_clip.SeparateFields().SelectEven()
bottom = interlaced_clip.SeparateFields().SelectOdd()

target_height_float = 243.104
target_height_int = 244
source_height_original = top.height
effective_source_height = target_height_int * source_height_original / target_height_float

crop_offset = source_height_original - effective_source_height + 1

top_resized = top.UserDefined2ResizeMT(640, target_height_int, b=0, src_height=effective_source_height, src_top=crop_offset)
bottom_resized = bottom.UserDefined2ResizeMT(640, target_height_int, b=0, src_height=effective_source_height, src_top=crop_offset + 1)
final_video = Interleave(top_resized, bottom_resized).Weave().AssumeTFF()

return AudioDub(final_video, audio)

The reason why halo not disapper is I need to plus 1 pixel in crop_offset.
so, top needed src_top=2, bottom needed src_top=1.
and exact target_height_float was 243.104, which was the value what I got in Premiere Pro.

I think I've done it...

[poisondeathray]

It's improved, but still some aliasing and halos . It makes you wonder if those are current settings are the ideal settings . Maybe the target height is still off, or the kernel resize settings are not ideal. A dead giveaway is the frame edges (top and bottom borders), and logo edges / halos such as around Toon!verse . Height - because you can also see corresponding halos around character top and bottom objects such as mouth, eyes etc.. not left vs. right halos - so not horizontal or width (or at least a lot less). If you can adjust the settings to looking at the logo as a guide, I suspect the result will be better

Another way for feedback/testing is to use avspmod tabs. You can put different script settings /version in different tabs flip tabs using the number keys quickly and see differences easily

[sdx12349]

It's improved, but still some aliasing and halos . It makes you wonder if those are current settings are the ideal settings . Maybe the target height is still off, or the kernel resize settings are not ideal. A dead giveaway is the frame edges (top and bottom borders), and logo edges / halos such as around Toon!verse . Height - because you can also see corresponding halos around character top and bottom objects such as mouth, eyes etc.. not left vs. right halos - so not horizontal or width (or at least a lot less). If you can adjust the settings to looking at the logo as a guide, I suspect the result will be better

Another way for feedback/testing is to use avspmod tabs. You can put different script settings /version in different tabs flip tabs using the number keys quickly and see differences easily

I don't care about Toon!verse logo. that is original 1080i logo, so It must have aliasing and halos.
and halo in frame edges and character also ok for me, because it is almost looking similar to original in my eye.