Kostya's Boring Codec World

As you can guess, my work on Real Video 4 encoder goes so well that I’d rather look at virtually anything else.

Probably some of you are aware of Tex Murphy game series (maybe not even just Mike) and I got reminded of an existence of The Pandora Directive game that supposedly has a lot of videos in it, so I downloaded a demo and decided to explore it a bit.

Game data is packed into extremely simple archive format (just the number of offsets and the offsets) so you need to know what archives to look into (hint: the larger ones).

So what about videos? They start with H2O magic (hence me calling it a watery format) and contain video frames and audio data (in an embedded WAV file). The video compression is both nothing special and somewhat original: it’s Huffman-compressed RLE. Video frame can contain several chunks, first there’s a Huffman tree description, then palette data and finally there’s compressed RLE data. The originality lies in the details: Huffman coding is often used to code single bytes while here single symbol represents both run length and value (also IIUC zero value signals skips), additionally the data is stored as the list of symbols for each codeword length. I don’t think I’ve seen anywhere neither RLE+Huffman used in such manner nor such tree description format.

This was a fun distraction even if I don’t care about the game series (I tried playing their other games like Countdown and Amazon: Guardians of Eden but the interface was too clunky and the action sequences were too annoying). It’s always nice to see some originality in video coding even if it is nothing advanced.

This is a resource format (as one could guess from its name) used by Beam Software for various resources in its games. The first time I had to deal with it was when some anonymous contributor contacted me with a REd decoder for videos in Lost Vikings 2. I adapted for FFmpeg (wow, it’s been fifteen years ago) and added sound support as well.

Recently I decided to revisit it (previously I tried to adapt decoder for 15-bit video but could not make it exactly right for some reason). So I looked at it, it turned out to be exactly the same algorithm but depending on whether palette is present in the frame (yes, 256-colour palette with RGB555 entries) pixel data is read either as byte palette indices or raw 16-bit values. I also looked at The Dame was Loaded game and its FCP format turned out to use exactly the same compression but with a different frame format (16-bit sizes, different audio/palette/video order and so on). And the game audio uses packed 12-bit PCM (two samples without low nibble are stored in three bytes) which is not something you often see.

And with that I’m done with game formats for a long time. Probably I should write a toy encoder or two now.

While the situation in Ukraine is stable bad so I keep looking at game formats instead of playing those games instead.

Apparently there’s a German adventure game from mid-1990s called Talisman and that’s about all I can tell about it. Yet it seems to have rather interesting animation format that’s worth talking about.

It turned out to be codebook-based compression with a codebook shared by a group of frames (typically 20-30) and actual codebook being created from a transmitted codebook data mixed with the static codebook (an interesting fact: unlike many other game formats, this one not merely allows seeking inside, there’s a special chunk preceding group of frames with the same codebook that contains an offset of the next such chunk). Tile operations are either copying 2×2 tile from another part of the frame (optionally updating one pixel in it) or reading it raw.

While it’s not the most complex format ever it turned out to be interesting enough. And with that I’ve probably exhausted the game video formats to look at. I’ll probably revisit SIFF but that’s about it.

While Iran wages a proxy war with Ukraine (and I can easily believe they’re not against doing it to the last russian), I’m still trying to distract myself by looking at various video formats in games I have no intention to play.

Finding games with unknown formats gets harder and harder (or at least more tedious). So here are two games with an interesting approach to video compression.

Dark Earth packs frames as separate planes with 5-bit components and each plane is compressed using a peculiar modification of LZ77 scheme: depending on a bit set, it either codes a run or a forward reference in either the same or previous buffer. Additionally it codes it using non-bytealigned stream (run mode is coded in 13 bits and copy mode takes 25 bits).

There’s also Defcon 5 game (the one from the mid-1990s, do not confuse it with a game with the same name but from a previous decade). This one employs a different method I don’t remember seeing anywhere else: painting squares. Of course you’re used to various codecs splitting tiles in blocks and filling them (from older codecs using quadtrees to H.26x and its rip-offs). This one does it differently though: there is a fixed set of tile sizes selected for each video (I’ve seen only 32, 16, 8, 4, 2 set though) and frame data is essentially a list of commands telling at which offsets to paint a tile of this size with a given colour (those offsets may be completely arbitrary). It may be not the most efficient compression algorithm but it’s quite interesting. Also the container specifies palette as fixed and changing colours—first part is stored in the header and the second part is transmitted with each frame. Again, it’s not something extremely effective but I can’t easily remember any other format doing it that way.

I’ll probably try to do something different next time but I consider time spent on looking at these formats well-wasted.

Since I have nothing better to do and I don’t feel a desire to write a lengthy rant about what’s wrong with this world (beside the fact that russia still exists of course), I keep looking at various games I can find in order to see if they have some interesting format. Last time I’ve managed to locate an elusive Indeo IVF sample, now it’s rather curious FMV format.

So there’s an unpublished PC game called Highlander: The Last of the MacLeods which is apparently some kind of 3D action game based on the animated series (one of those, the franchise seems to have a lot in it beside the one good movie). And it turns out that it has a three dozen of FMV clips in proprietary format (stored inside mov.dat but extracting them is trivial).

From technical point of view the format is rather simple but video compression is rather peculiar: each frame in coded independently by painting each 4×4 block with 1-16 pixels using a certain pattern. That’s common but this data is packed with a simple history-based method: depending on a flag a value is rather read raw or reused from a history buffer depending on two previously decoded values. And now there’s a twist—this compression is applied twice.

Overall, it was fun to look at the format even if I don’t care about the game itself and the video quality is rather poor (probably thanks to the rather low bitrate and fixed palette all those FMVs share despite half of them being digitised animation clips and half of them being recorded in-game 3D animation). Still, now I can play the files with NihAV if I ever get such desire.

It’ll take a long time to liberate Ukraine from the invaders, so while it goes on I have to entertain myself somehow.

Since I had nothing better to do I looked at two FMV games—Deadly Tide and Hot Wheels: Stunt Track Driver. Both of them have their levels packed as video data with some additional control data.

Deadly Tide is more complex one with its L96 format: it defines a resource file with several possible sub-levels and three different video streams (something like 320×240, 160×120 and 80×80) compressed with Cinepak plus half a dozen of various control resources and audio stream.

Hot Wheels uses JAM format which is a series of compressed chunks. 0 – frame data, 2 – palette update (strangely enough it updates palette not from the start but rather to the end), 3 – some control data. Frames are stored either unpacked, packed with LZSS or packed with differential LZSS (in this case copy operation takes bytes from the previous frame). Oh, and it uses obfuscated WAV files for audio that I didn’t bother to look at.

I’m still looking for an interesting format to RE, meanwhile those two were at least somewhat interesting.

Recently I’ve encountered a review of an old FMV game called Ripper and my professional interest was to look what formats it uses. Well, it turned out to use Smacker and AVI. And those AVI files are their own format that uses Smacker in perverse way: it stores a sequence of commands with 14-byte header that control video playback. The first dozen of commands are usually “fill audio buffer with this data” commands and there’s one command for initialising video that includes full Smacker header inside its payload. From what I see it abuses Smacker decoder by constructing file on the fly using the data from the commands (I’m pretty sure the following commands have video frames in their payload).

I’m not sure if I’d want to pursue it further and document it let alone adding support for it in NihAV but that format was definitely something.