Q?: Is Theory still improving much? Is it possible to improve it a lot without breaking backward compatibility, or would that require breaking legacy support?

Q?: Is Theory still improving much? Is it possible to improve it a lot without breaking backward compatibility, or would that require breaking legacy support? The encoder still has a good bit of room to improve; it still lacks proper pixel-domain RD optimization and could use some psychovisual improvements as well. It is primarily crippled by the ancient VP3 specification though. In theory it may eventually be able to eke ahead of Xvid, as while Xvid was a very good MPEG-4 ASP encoder, it never had any significant psy optimizations, so even though Theora as a video format is probably slightly worse, the encoder should be able to surpass Xvid with psy optimizations in some cases. The primary problem for adoption (not for quality) is its lack of features. For many many use-cases, one needs fine-grained control over various parameters, such as the buffer size, the maximum bitrate, the latency, and so forth. The encoder is currently very inflexible, and with no alternative encoding libraries, this leaves very little choice for many applications. For practical purposes, it only has one speed mode: "slow as hell", which makes it basically impossible to stream anything above SD.