All discussion of http://eye.swfchan.com/services/swfh264/ goes here.

Trivia: SwfH264 took exactly 10 days to develop.

FLV files created in FLV Maker are encoded using VP6 + Mp3. These can be embedded into the main timeline of a swf, having the benefit of being seekable to every frame and being streamed (you can start watching the video before downloading the entire flash).

FLV files created for SwfH264 are encoded using H.264 + AAC (newer codecs). Those will produce much better quality than the old VP6/Mp3 combo but cannot be embedded into the main timeline of a swf. Flashes created from SwfH264 simulates that the FLV is being downloaded from somewhere else and due to how byte arrays are embedded into swf files the flash must be fully downloaded before the video can start to play.

If you use a VP6/Mp3 kind of FLV with SwfH264 there will be no increase in quality per byte, the streaming feature will just be sacrificed for nothing. Also, FLVs that are created with FLV Maker must still have their metadata fixed or else they won't be seekable in the flash created with SwfH264 (the reason that they are seekable in flashes created with the toolkit in FLV Maker is because they are able to be embedded on the main timeline - and then they are more accurately seekable because they don't rely on keyframes in the video).

If you discover any bugs or anything else that looks weird please mention it here.

Fun little app to play with. I posted my experiment with it here: http://swfchan.org/7231/

For now I recommend that you always create the subtitles using a proper .srt program. It's mentioned on the project page as well in the example (next to the third screenshot).

I didn't want to go overboard with the initial features in SwfH264, as with most of what I do I try not to waste too much time on something until I see it actually being used by people. Further development of SwfH264 depends on if I start seeing many flashes created by it, or if people demonstrate interest in the program by posting here.

How in the hell can vector fonts be that large? Did they include every Unicode code point in existence?

Or just:
java.awt.GraphicsEnvironment.getLocalGraphicsEnvi ronment().getAllFonts()
to get every font installed on the system, regardless of platform.

The default behavior of loading a font in Java should be cross-platform, as far as I know it always searches all available locations for the entered font name and should find it as long as it's installed.

I tried to decompile it and managed to build a working version. Unfortunately video/subtitles embedded data was loaded in frame 1 (instead of frame3) meaning that preloader didn't work at all. My Flash/AS3 knowledge is too weak to fix that.

Flashbulb's here if you want to try: www.xeloh.com

I seriously need to stop replying to myself.

>Unfortunately video/subtitles embedded data was loaded in frame 1 (instead of frame3) meaning that preloader didn't work at all.
With the .fla project open go to "File" > "ActionScript Settings..." to open the Advanced ActionScript 3.0 Settings window. There you change "Export classes in frame" from 1 to 3. Perhaps that will get your hack to work?

>>21211
C'mon, do the right thing and release sourcecode. No point in keeping progression of this app in the gutter. Once the source is released, we can get on about fixing the loose ends and wrapping it in a proper WYSIWYG interface for those who don't into Adobe+Sony. (parametric controls are for linuxfhags, not humanz)

>>23794
I'll probably include less fonts in the next version to be honest. It's because you can use any font that is installed on your computer, go to Control Panel > Fonts to browse everything you can use in SwfH264.

However if you have a suggestion for a super good font that people usually don't have installed by default in Windows please let me know!

The biggest issue is that 4:3 videos that look great at 360p are expanded by default to a great size, even larger than my web browser window.

If it would possible to just have a patched release that outputs smaller defaults, that would be fine also.

Its just kind of ugly when a low bitrate >360p video is enlarged to nearly the entire browser window.

Also a tip about the AAC codec: http://wiki.hydrogenaud.io/index.php?title=Advanc ed_Audio_Coding#Encoders_.2F_Decoders_.28Supported _Platforms.29
Check the table on the right. Not trying to push you toward, just giving you the information I've been told, in case you were wondering ever at some point. I personally use FDK.

Have a few notes on what to change in the next version of SwfH264. Here's all the changes so far:

#######################

ffmpeg.exe -ss 00:00:00.000 -i "input.mp4" -vb 555k -t 99:00:00.000 -r 18.0 -vf crop=iw-0000*2:ih-0000*2,scale=-1:ih*1.00 -keyint_min 9 -g 180 -pass 1 -passlogfile pinfo_delme -tune grain -pix_fmt yuv420p -profile:v high -level 5.2 -preset veryslow -max_interleave_delta 0 -vcodec libx264 -sn -acodec aac -strict experimental -ac 2 -ar 44100 -ab 64k -cutoff 18000 -f flv -y NUL
ffmpeg.exe -ss 00:00:00.000 -i "input.mp4" -vb 555k -t 99:00:00.000 -r 18.0 -vf crop=iw-0000*2:ih-0000*2,scale=-1:ih*1.00 -keyint_min 9 -g 180 -pass 2 -passlogfile pinfo_delme -tune grain -pix_fmt yuv420p -profile:v high -level 5.2 -preset veryslow -max_interleave_delta 0 -vcodec libx264 -sn -acodec aac -strict experimental -ac 2 -ar 44100 -ab 64k -cutoff 18000 -f flv -y "output.flv"

#######################

Note that this might change when the time comes. Especially "-tune grain", which wants the input video to be of excellent quality. "-acodec libvo_aacenc" was replaced with "-acodec aac -strict experimental" because the native experimental AAC encoder is supposed to give better quality than libvo_aacenc from what I've read. I'd like to make the new acodec "libfdk_aac" instead (for Fraunhofer AAC encoder) if I can get a compiled version of ffmpeg.exe with the codec installed. Don't know if Apple AAC is possible? That sounds very licence-esque.

The bundled general-solution will be 2-pass since that improves quality further, even if that means changing stuff on two lines. Question is if it always improves it or if it could make it worse sometimes? I actually did try 2-pass out for the first version of SwfH264 but I must have fucked it up because I didn't notice any quality improvement and decided to skip it for simplicity's sake. But since then I've noticed that yes of course it improves the final result. If someone sees this that happens to be an expert on making 2-pass encodings with ffmpeg please post any advice that you can! Maybe there's more to it than I've written in this post?

Two-pass encoding is used when you want a consistent average video quality throughout a video at a specific resulting file size. What is really going on behind the scenes is on the first pass, ffmpeg analyzes the file and builds a data structure that indicates where more bits are needed than others to maintain an average q level throughout. Then on the second pass, that analysis is used to vary the bitrate as necessary to reach closer to a specific file size.

This is nice because if you want a file that is 10MiB in size, it is much easier to get a result close to that with two pass encoding. You divide the file size (as kbit) by duration (in seconds) then subtract the audio bitrate and then subtract maybe 2% of the resulting value to account for overhead and you get an appropriate value for where you would want your bitrate to be.

So for example, you have a 60 second video, you know your audio is going to be 32kbit/s. Take 10 MiB * 1024 = 10240 KiB * 8 = 81920 Kbit / 60 secs = 1365.3 Kbit/s - 32 Kbit/s = 1333.3 - 2% = 1306 Kbit/s approx.

What you are seeing by using 2 pass encoding is that certain scenes that would look much better take a quality hit so that more complex scenes can maintain consistent quality with the less complex scenes. For /f/ and the like, this is acceptable.

>if I can get a compiled version of ffmpeg.exe with the codec installed. Don't know if Apple AAC is possible? That sounds very licence-esque.

You can't legally distribute ffmpeg with libfdk_aac built in. That changes its license to non-free. However, it would be very beneficial to indicate an easy solution to this because libfdk_aac supports high efficiency AAC (aac-he). AAC-HE can be used at low bitrates (think 32k and lower) and still sound passable. I usually use it around 32K because the native aac encoder doesn't even sound comparable at 64k with normal aac.

Even 64k sounds better

Examples:

64k AAC-HE
http://eye.swfchan.com/flash.asp?id=180435&n=Weir d+Faced+Umi-chan%27s+Munchy+Monk.swf

32k AAC-HE
http://swfchan.com/37/181257/?ME!ME!ME!+Full+HQ.s wf

You could nudge people to use this to autobuild a static copy of ffmpeg. Just *don't* build mediainfo or it bombs out, and its not necessary to build all the extra binaries so long as you just want ffmpeg.

https://github.com/jb-alvarado/media-autobuild_su ite

There's been some pushback on /f/ because the average joe doesn't look further than the general tool, and so the results have been poor. I think the options you've got now will help a fair bit but IMO anything that can be done to take tweaking steps out of the hands of average joe would help the appeal of this tool.

This was the encoding settings I performed:
ffmpeg -i jkjtxf.flv -acodec copy -preset placebo -threads 1 -b:v 110k -subq 10 -sc_threshold 10 -bf 16 -b_strategy 1 -x264opts aq-mode=2:merange=100:direct=auto:partitions=all: 8x8dct:qpstep=7:frameref=16:fast_pskip=0 -vcodec libx264 -tune animation -qcomp 1 -refs 16 -me_method full -threads 1 -trellis 2 -qmin 30 -vf scale=640:360 -pass 1 kog.flv

The audio is 32k AAC_HEv2

If you use the animation tune you don't really need to set the other things, like trellis and the like. This would work just about as well, but you can try adding on options slowly if you want after.

ffmpeg.exe -y -i jkjtxf.flv -vb 110k -vcodec libx264 -sn -acodec libfdk_aac -profile:a aac_he -ac 2 -ar 44100 -ab 32k -cutoff 18000 -pass 1 -tune animation -profile:v high -preset veryslow -f flv dump.flv
ffmpeg.exe -y -i jkjtxf.flv -vb 110k -vcodec libx264 -sn -acodec libfdk_aac -profile:a aac_he -ac 2 -ar 44100 -ab 32k -cutoff 18000 -pass 2 -tune animation -profile:v high -preset veryslow -f flv output2pass.flv

Run both in that order. Delete dump.flv afterwards or if using linux change dump.flv to /dev/null

If your ffmpeg build does not have libfdk_aac then you'll have to use the native encoder most likely or build ffmpeg with libfdk so you can use AAC-HE.

If your audio source is 32k however, you might try to get a higher quality source video if possible.

Anyway, while we are talking about features, I would request the ability to loop back to a previous time. Like
>Video Starts from 0:00
>Video hits 1:24 that repeats at 0:54
>Video loops 0:54 to 1:24 indefinitely

I have thought up a workaround that involves loading chunks of movie bytes as jpg images, which don't have to all be loaded on the same frame. You basically have to trick flash into thinking it's loading a jpg when it's really just an array of bytes. When the fake image is loaded you extract the bytes from it and append it to the main byte array of the movie. Storing the bytes as audio data would probably be better or even easier than doing it as fake images... Flash is great at streaming audio chunks from the main timeline and those are expected to contain all kinds of random bytes. Anyway, haven't put it into practice yet but in theory it should be possible.

While I'm rambling, wonder if Adobe will add H.265 support to flash? That would basically mean twice the quality for the same file size, judging from what I've seen of the codec so far. If nothing else it would mean that Flash keep up with the times.

Another problem with that codec is it is the available encoders are rather difficult to tune correctly such that they were compete well on quality with H.264. This will need further attention to rectify.

Fug, I mean H.265 can't really compete with H.264 well on quality at any rate without significantly more work outside of benchmarks.