A whole film on one disc, and a codec that was never meant to ship

The codec is the argument · Long

A hacked Microsoft encoder, a single CD-R, and the accidental birth of the consumer video codec.

The hack that opened the door

In late 1999, a feature film fitting on a single 700 MB CD-R was not a solved problem — it was closer to a provocation. DVD had established that high-quality digital video was possible, but the disc was large and the hardware expensive. CD-R blanks, by contrast, cost almost nothing and played in every PC with a burner. The gap between those two facts was where Jérôme Rota worked.

What Rota produced was DivX ;-) 3.11 Alpha — note the winking emoticon, a deliberate signal that this was not a product. It was a modified build of Microsoft's own MPEG-4 Video codec, part of the NetShow / Windows Media lineage that Microsoft had developed as a streaming format and never intended for high-bitrate offline use. The modification stripped out the rate-control logic that kept the codec inside the narrow bandwidth assumptions of a streaming session, and replaced it with something that would simply use whatever bitrate the encoder was given. The result was a pre-standard MPEG-4 codec operating well outside its original design constraints — and producing, for the first time in a broadly accessible form, video quality that could survive a feature-length encode and still fit on a disc most people already owned.

The codec borrowed the name "DivX" partly as a gibe at an earlier, unrelated pay-per-view disc scheme called DIVX — a Circuit City venture that had collapsed in 1999 after failing to persuade consumers that a disc that expired was a good idea. The joke was embedded in the name from the start.

From hack to format

The 3.11 Alpha label mattered: it was never presented as finished or legitimate. What it demonstrated, though, was a genuine capability gap that the established industry had not addressed. MPEG-2, the codec underpinning DVD-Video, produced excellent quality but at bitrates that assumed a 4.7 GB disc and dedicated hardware decoders. Nothing in the consumer software space had shown that a software-decoded, broadly compatible MPEG-4 codec could compress a two-hour film to under a gigabyte without the result looking like it had been filmed through frosted glass.

The DivX ;-) releases demonstrated exactly that, and the compression mechanism was orthodox enough to explain: MPEG-4 Part 2 uses discrete cosine transforms applied to 8×8 pixel blocks, inter-frame prediction to describe most frames as differences from earlier reference frames, and variable bitrate allocation to concentrate detail where the image changes most. The hack did not invent any of these tools. What it did was route a mature codec's internals toward a different use case — offline, high-bitrate, software-decoded — and prove the combination worked on hardware people already had.

DivXNetworks formed in the aftermath to build a legitimate, commercially licensable product on top of the interest the hack had generated. DivX 4 (2001) and DivX 5 (2002) were clean implementations that added two-pass encoding — a mode where the encoder made a first pass to analyse the whole film before making a second pass to allocate bits more intelligently — along with proper B-frame support and psychovisual optimisations. The codec became a commercial entity with a licensing structure and, eventually, hardware support in standalone DVD players. What had begun as a curiosity in a winking alpha build was, within three years, a format that consumer electronics manufacturers were paying to implement.

The container problem nobody could ignore

All of this video had to live somewhere, and for most of the DivX era that container was AVI — the Audio Video Interleave format Microsoft had introduced in 1992. AVI was not designed for variable bitrate streams, had no native support for modern audio formats at the bitrates users actually wanted, and its index structure assumed file sizes that CD-R rips regularly violated. Encoders and players worked around these limitations through a series of OpenDML extensions and hacks of their own, but the seams showed. Files that played perfectly in one player had sync drift in another; audio and video could desynchronise over the length of a film if the rate-control assumptions embedded in the container did not match what the codec had actually done.

The tools that handled this material on the editing and re-encoding side were equally improvised. Avery Lee's VirtualDub, a frameserver and encoder front-end he had written and released freely, became the dominant way to process DivX-era AVI files — not because it was officially supported, but because it handled the format's edge cases more reliably than anything else available.

What the codec cost and what it established

DivX 5 introduced the codec's most contested commercial decision: the encoder required payment for the nag screen to disappear, creating the first widely felt moment where a codec's business model became part of its user experience. The decoder remained free. This asymmetry — free to play, pay to encode — would recur in later licensing debates. By then XviD, an open-source MPEG-4 Part 2 implementation, had already forked from the OpenDivX code to provide a free-as-in-freedom alternative to the commercialised DivX encoder.

The Moving Picture Experts Group had standardised MPEG-4 Part 2 visual coding in 1999, and both DivX and its open-source counterpart were implementations of that standard — meaning the underlying patent portfolio, administered by MPEG LA, applied regardless of which encoder produced the file. Royalty-free was not on the table for any MPEG-4 Part 2 implementation; the question was always which entity collected the fee and on what terms.

What the DivX line established, beyond its own commercial trajectory, was proof of concept for the consumer video download era. It demonstrated that a software codec, running on consumer hardware, could produce acceptable quality at bitrates low enough for physical distribution and, in time, for network transfer. H.264, standardised in 2003, arrived with substantially better compression efficiency and eventually displaced MPEG-4 Part 2 in every serious application — but its consumer acceptance was built on a public that had already learned, through a hacked alpha build and its successors, that compressed digital video on a computer was a normal thing to watch.