📜Context Length

A potentially very powerful tool you can use to keep your song consistent or suddenly shake it up!

This is a very powerful extending tool. It is essentially how much of your current song (within the confines of what you are cropping) the AI can "see".

This is important because the AI refers to the song pieces it can see to generate the new clips, so the song remains consistent, progresses and builds upon elements. This is what keeps things like the instrumental setup and vocalist consistent. If it couldn't do this, your songs would be totally random and change every extension! Of course, perhaps that is what you want...that's up to you to decide!

Context length is a simple slider ranging from 1 second of context, to 130 seconds.

The control is a single slider which specifies how much of the song the AI can refer to when generating the clip ranging from 1 second up to 130 seconds.

Setting it to 1 second means the AI can only "see" the last second of your song which means it has very little context to work off, and doesn't "know" what the song was doing before that point, so is unlikely to replicate it.

Sliding it up to 130 seconds means the AI will refer to all of that length and knows "what's been going on" for much of the song. So certain themes, melodies and such will reappear.

Note: When extending after, the context is counted backwards from the point your generation would start from. When generating before, the song context is counted forwards from the point your generated clip would end.

Generally speaking the more consistent you want a song overall, the higher the slider should be this is particularly true if you want returns to melodies that the song covered much earlier (so long as it is within 130 seconds). Higher values however reduce the chance for "sudden changes" because the AI will prioritise the coherence of the clip.

The less consistent you want your clip, the lower the slider should go, the less the AI can "see", the more it has to be creative to finish the generation, which means it will deviate from the wider context of the clip. If you want to avoid an earlier part of the song sound-wise, slide this lower.

Note: Although higher slider context makes a song more consistent, it does not need to be extremely high to keep certain elements of a song. For example it is possible to "keep" the same vocalist within a song and the same basic instrumental elements with as little as 4 seconds of context.

You can forcefully push this slider low when you want a sudden change, including changing a vocalist in a song. This is part of how you can easily invoke duets within a song, by causing the AI model to be unable to "see" the current vocalist, so it will generate a new one instead! After this you can widen the context so it can "see" both and it can begin to duet them properly (with proper prompting).

There are many tricks and cool sounds possible through creative use of this tool, it is definitely one of the more powerful ones in the Udio toolkit, when used properly.

Note: Sliding context low does not affect cropping. So if you specify the AI should only use 5 seconds of context, but there is 32 seconds prior, all 32 seconds of that will be preserved in the next generation, but only the last 5 of it will be used to inform that clip!