anime encoding

[|SeNsUi|]

Hi.. I jst got a problem on encoding a anime series. In dark areas of the anime i can see some sort rainbows colors and moving pixels. its like that on the DVD as well. what filters would i need to use to minimize this any ideas?

[Chiba]

Screens ?

[Nightshiver]

Always post a video sample.....
Beyond that, if there actually are "rainbows" in your original DVD, then you simply need a de-rainbow script. There are several methods for doing this. Use search.

[thewebchat]

$5 says that Mr. "|SeNsUi|" (lol) won't be returning, but:

"rainbow colors": Now, this could just be rainbowing, but since he says it's in "dark areas", my guess is that it's banding from MPEG fail. In which case, the solution is to add lots of noise. If it's actually rainbowing, the solution is to blur the crap out of the chroma.

"moving pixels": Could be blocks (?), dotcrawl (unlikely), grain/noise (?), something else (?). The solution to all of these is to blur the crap out of the luma (and chroma if necessary).

[|SeNsUi|]

sorry for the late reply. iv been searching the forums iv tried the de rainbow but to no avail. nw i shall try adding noise as it luks like mpeg fail.. here r the screen shots

[Ghitulescu]

sorry for the late reply. iv been searching the forums iv tried the de rainbow but to no avail. nw i shall try adding noise as it luks like mpeg fail.. here r the screen shots

The left image seems to me a de/interlacing problem. Anime are well known for mixing various fps (and interlaced/deinterlaced sequences) within the same clip.

[J_Darnley]

Banding, not rainbowing, is the problem here, apart from what Ghitulescu pointed out. Use gradfun2db, gradfun2dbmod, other noise/grain generators.

[|SeNsUi|]

I used megui to analyze the video then it spat out the best deinterlacing method, is this not a good way to do it? I am downloading the grad files and shall test em out.

thank you for all the help

[thewebchat]

The "best" way to deinterlace it would be to actually *look at it* instead of using some dumb script (lol megui). Anyway, in screenshot 1, I see a field blend. This is probably due to your ivtc not finding a match and deciding to blend two fields together. You can make it interpolate instead which might look a bit better. On his shirt and on the wall, the problem you are seeing is not rainbowing but banding. Add some noise. On the second screenshot, what you're seeing is blocking on his shirt. Blur it out with dfttest, fft3d, deblock_qed, or some other blurring filter. Alternatively, you could just add even more noise and it'll cover that up too.

[|SeNsUi|]

the first screenshot i posted the clip was playing i never paused it and tookk screen shot which is why it looks like that. cud sum1 give me a good setting for grad.. i already use fft3d but im still a newbie dunno the best to add noise.

my setting right now is

FFT3dFilter(sigma=1.5, bt=3, bh=50,bw=50, sharpen=0.5, measure=false, degrid=1)

[thewebchat]

the first screenshot i posted the clip was playing i never paused it and tookk screen shot which is why it looks like that. cud sum1 give me a good setting for grad.. i already use fft3d but im still a newbie dunno the best to add noise.

my setting right now is

FFT3dFilter(sigma=1.5, bt=3, bh=50,bw=50, sharpen=0.5, measure=false, degrid=1)

gradfun2db takes no parameters. Just run it as "GradFun2DB()". If you need more, just paste it a few more times.

[|SeNsUi|]

hmm nice filter this is itt has minimized the banding radically..

I ran both debaning filters, after the denoising and sharpening.

GradFun2DBmod()
GradFun2DB()

[|SeNsUi|]

wicked the clip is perfect :)..btw is deblock_qed good or should i stick to x264 deblocking.. iv got the avs script and everything mite give it a try.

[J_Darnley]

wicked the clip is perfect :)..btw is deblock_qed good or should i stick to x264 deblocking.. iv got the avs script and everything mite give it a try.
x264 doesn't try to remove blocks in the input video. The H.264 in-loop filter tries to prevent blocks forming in the decoded video.

gradfun2db takes no parameters. Just run it as "GradFun2DB()". If you need more, just paste it a few more times.
gradfun2db() takes a float argument that controls its strength. The default is 1.2 IIRC

[thewebchat]

x264 doesn't try to remove blocks in the input video. The H.264 in-loop filter tries to prevent blocks forming in the decoded video.


gradfun2db() takes a float argument that controls its strength. The default is 1.2 IIRC

The thing is, you can only really raise it to 1.000001 since if it's <=1, strange artifacts appear everywhere. And gf2db at 1.000001 is not much stronger than 1.2, which is already not very strong.

[J_Darnley]

How can you raise 1.2 to 1.001? And I don't know what you mean by "strange artifacts appear everywhere". I haven't seen any problems (other than the non-processing of edges). With a high strength you get a blurring effect.

[thewebchat]

How can you raise 1.2 to 1.001? And I don't know what you mean by "strange artifacts appear everywhere". I haven't seen any problems (other than the non-processing of edges). With a high strength you get a blurring effect.

gf2db gets weaker as you increase thr

[|SeNsUi|]

thanks for all the help this has improved the issues i had :).