Smallest PNG File Size, Part 3

[Jay123210599]

I found that VirtualDub2, Shotcut, and Kdenlive can make png files smaller than ffmpeg. How do I adjust the settings to match the file size (or make it even smaller, if possible)?

Also, can ffmpeg make zip/7z/rar/zqad files?

[poisondeathray]

Use the ffmpeg png -pred (prediction method) settings listed in the help file to adjust the ffmpeg png compression . Default is "none" so it will produce larger files but export faster than say, vdub2's or shotcut's or kdenlive's png's . Higher compression ratio means slower exports as usual for tradeoffs

[Jay123210599]

And my other question?

Oh, I tried what you said but they're still not smaller.

[poisondeathray]

Don't forget about specifying proper pixel type - you might be exporting 16bit RGB . By default if input video is >8bit, ffmpeg will export 16bit PNG's

Don't forget about proper YUV to RGB conversion or you'll get wrong colors

Don't forget the warning in your other thread about colorimetry flagged input YUV video to PNG sequence using ffmpeg - and the cHRM, gAMA tags - or PNGs will display differently in different programs and browser versions

[poisondeathray]

Also, can ffmpeg make zip/7z/rar/zqad files?

I think not. They are not listed in the formats help

[Jay123210599]

Don't forget about specifying proper pixel type - you might be exporting 16bit RGB . By default if input video is >8bit, ffmpeg will export 16bit PNG's

Don't forget about proper YUV to RGB conversion or you'll get wrong colors

Don't forget the warning in your other thread about colorimetry flagged input YUV video to PNG sequence using ffmpeg - and the cHRM, gAMA tags - or PNGs will display differently in different programs and browser versions

My videos are 10bit.

[poisondeathray]

So specify 8bit PNG. Most displays will be 8bit RGB anyways. There is no benefit to use 16bit RGB for posting to a website for display, it will just make filesize larger

You know how to use zscale, format filters from the other threads . Sometimes ffmpeg gets it correct "automatically", sometimes not . It depends on the source video, flags, and the ffmpeg version

But you will get the wrong colors in some browsers like firefox with that workflow, because of the cHRM, gAMA tags if YUV video is colorimetry flagged . If using ffmpeg to export PNG's , I would use an avisynth script with PropClearAll() . That clears the flags and the exported PNG's will not have those tags, colors will look correct everywhere. You can also control the YUV to RGB conversion properly in avisynth if you wanted to

Vdub2 will look at flags for it's RGB conversion by default. If video flags are correct, RGB conversion should be correct. If flags are wrong, conversion will be wrong. If video is unflagged, it will use Rec601 for RGB conversion - it will be wrong for most HD sources

[Jay123210599]

So specify 8bit PNG. Most displays will be 8bit RGB anyways. There is no benefit to use 16bit RGB for posting to a website for display, it will just make filesize larger

You know how to use zscale, format filters from the other threads . Sometimes ffmpeg gets it correct "automatically", sometimes not . It depends on the source video, flags, and the ffmpeg version

But you will get the wrong colors in some browsers like firefox with that workflow, because of the cHRM, gAMA tags if YUV video is colorimetry flagged . If using ffmpeg to export PNG's , I would use an avisynth script with PropClearAll() . That clears the flags and the exported PNG's will not have those tags, colors will look correct everywhere. You can also control the YUV to RGB conversion properly in avisynth if you wanted to

Vdub2 will look at flags for it's RGB conversion by default. If video flags are correct, RGB conversion should be correct. If flags are wrong, conversion will be wrong. If video is unflagged, it will use Rec601 for RGB conversion - it will be wrong for most HD sources

What about be the avisynth script or how do I use it to make sure that all colors are correct for png files and well as lossless webp and jxl files, and get low file sizes for them?

[poisondeathray]

What about be the avisynth script or how do I use it to make sure that all colors are correct for png files and well as lossless webp and jxl files, and get low file sizes for them?

It depends on the input video and output format

There is no 1 size fits all.

Different videos have different pixel types, different matrix/transfer/primaries , HDR vs. SDR . If you use the same settings for everything, you're going to get the wrong result for many cases . You saw an example of that before in your other threads

So it might be a different script for each video, same with output format. I only suggested the script as a workaround for ffmpeg png's . The ffmpeg png issues has reported many times . Many software that use ffmpeg for png exports have the same problem

For the compression side, use the highest compression settings in the help file and it will be the slowest as well. Trade offs.

[Jay123210599]

What about be the avisynth script or how do I use it to make sure that all colors are correct for png files and well as lossless webp and jxl files, and get low file sizes for them?

It depends on the input video and output format

There is no 1 size fits all.

Different videos have different pixel types, different matrix/transfer/primaries , HDR vs. SDR . If you use the same settings for everything, you're going to get the wrong result for many cases . You saw an example of that before in your other threads

So it might be a different script for each video, same with output format. I only suggested the script as a workaround for ffmpeg png's . The ffmpeg png issues has reported many times . Many software that use ffmpeg for png exports have the same problem

For the compression side, use the highest compression settings in the help file and it will be the slowest as well. Trade offs.

I looked at the details of the png files and it turns out the ffmpeg png files are 48 bit while the other pngs are 24 bit. What is and how do I get the lowest bit for png files, and will it affect the quality of the images?

[poisondeathray]

I looked at the details of the png files and it turns out the ffmpeg png files are 48 bit while the other pngs are 24 bit. What is and how do I get the lowest bit for png files, and will it affect the quality of the images?

"24bpp" or "bit per pixel" is the same thing as 8bpc or "bit per channel" . 8bit Red, 8bit Green, 8bit Blue

-vf format=rgb24

Yes lower quality; but on a 8bit RGB display, you will not see any difference - and there will be no additional compression losses eitherway (you're not going to get those additional artifacts such as in the jpeg or jpegxl thread)

The 8bit RGB image will be lower quality if your source material was true 10bit or higher (not 10bit encoded from a 8bit source) . And on a 10bit display (there are getting more common these days, but the majority are still 8bit), you might see a difference in quality, especially around gradients. The 8bit RGB image will show more banding

[Jay123210599]

I looked at the details of the png files and it turns out the ffmpeg png files are 48 bit while the other pngs are 24 bit. What is and how do I get the lowest bit for png files, and will it affect the quality of the images?

"24bpp" or "bit per pixel" is the same thing as 8bpc or "bit per channel" . 8bit Red, 8bit Green, 8bit Blue

-vf format=rgb24

Yes lower quality; but on a 8bit RGB display, you will not see any difference - and there will be no additional compression losses eitherway (you're not going to get those additional artifacts such as in the jpeg or jpegxl thread)

The 8bit RGB image will be lower quality if your source material was true 10bit or higher (not 10bit encoded from a 8bit source) . And on a 10bit display (there are getting more common these days, but the majority are still 8bit), you might see a difference in quality, especially around gradients. The 8bit RGB image will show more banding

How about this? How do I get ffmpeg to take screenshots (not extract frames) of every single frame in a video?

[poisondeathray]

How do I get ffmpeg to take screenshots (not extract frames) of every single frame in a video?

Screenshots of every single frame in a video - is the same thing as exporting an image sequence from a video

Or maybe you're not describing what you want correctly ?


(and if video is YUV, and tagged with colorimetry flags - the output PNG sequence will have cHRM, gAMA tags that cause the PNG to display with wrong colors in firefox, chrome, edge(chromium) - thats about 90% of browsers)

[Jay123210599]

How do I get ffmpeg to take screenshots (not extract frames) of every single frame in a video?

Screenshots of every single frame in a video - is the same thing as exporting an image sequence from a video

Or maybe you're not describing what you want correctly ?


(and if video is YUV, and tagged with colorimetry flags - the output PNG sequence will have cHRM, gAMA tags that cause the PNG to display with wrong colors in firefox, chrome, edge(chromium) - thats about 90% of browsers)

I mean like actually taking a screenshot on my computer (e.g. I press Win+PrtScr to take a screenshot and it produces a png file in kbs).

[poisondeathray]

I mean like actually taking a screenshot on my computer (e.g. I press Win+PrtScr to take a screenshot and it produces a png file in kbs).

ffmpeg does not display video . It's not a video player.

If you mean screen capture the video displayed from playing in another application - there will likely be problems like dropped frames

[Jay123210599]

I mean like actually taking a screenshot on my computer (e.g. I press Win+PrtScr to take a screenshot and it produces a png file in kbs).

ffmpeg does not display video . It's not a video player.

If you mean screen capture the video displayed from playing in another application - there will likely be problems like dropped frames

What if I use -f image2 and/or -frames:v with making an image sequence of png files?

[poisondeathray]

What if I use -f image2 and/or -frames:v with making an image sequence of png files?

-f image2 specifies the format as an image sequence, I don't see it helping in any way

-frames:v "n" limits the number of written images. eg. If "n" was 100, it would only write 100 images

[Jay123210599]

What if I use -f image2 and/or -frames:v with making an image sequence of png files?

-f image2 specifies the format as an image sequence, I don't see it helping in any way

-frames:v "n" limits the number of written images. eg. If "n" was 100, it would only write 100 images

So my only option to get images sequences of kbs is using lossy compression?

[poisondeathray]

So my only option to get images sequences of kbs is using lossy compression?

Depends on the source

But for the examples you posted - probably

You can crop to a region of interest at 1:1 . e.g. display the top left corner.


For websites that have limits on image upload filesizes, you can post links to 3rd party image hosts . This is a very common practice