Blend: A Playlist Built From Where Two Tastes Overlap

A blend is not a merge. Dropping two libraries into one playlist gives you a playlist neither person wants. Here is what Blend actually does instead.

Published Updated

The naive approach to a shared playlist is to interleave two people's top tracks. It reliably produces something both people tolerate and neither enjoys, because it is two playlists taking turns rather than one playlist.

Blend takes a different approach: work out where the two histories actually agree, and build outward from there.

How similarity is scored

Each member gets a vector describing their listening — which artists they play and how much, and which genres, drawn from recent history rather than all time. Comparing those vectors gives a similarity score.

Artist overlap counts for more than genre overlap, and deliberately so. Two people who both listen heavily to the same specific artists have far more in common than two people who both nominally listen to 'pop', which describes almost nobody usefully.

How the playlist is built

The generated playlist is not a ranked list of everyone's top songs. It mixes several kinds of track deliberately:

Shared tracks
Music both members already play. These are the anchors — the points where the playlist is guaranteed to work for everyone.
Bridge tracks
Music that sits between the two tastes rather than inside either. This is where a blend earns its keep, and it is the part a manual shared playlist never produces.
Individual favourites
Tracks from each member's own listening and liked music, weighted so the playlist still sounds like both people rather than like an average of them.

There is also a sequencing rule: the generator avoids stacking several consecutive tracks from the same contributor, so the playlist keeps alternating rather than settling into one person's block. Tracks carry attribution, so you can see who each one came from.

It keeps updating

A blend is not generated once. It refreshes on a schedule as members keep listening, so a playlist made in March reflects what everyone is playing now rather than what they were playing then.

There is also a lighter, faster signal: what members are listening to right now nudges the blend slightly, within a bounded limit so a single evening cannot take the whole thing over.

Groups, not just pairs

Blends are not limited to two people — a blend supports a small group, up to ten members. The dynamics change as the group grows: with two people the shared core is usually thin and the bridges do the work; with six, there is normally a genuine common core and the individual sections shrink.

Using it well

  1. Create the blend and then forget about it for a fortnight. It gets better as it accumulates history from everyone in it.
  2. Do not curate it manually. Removing everything you did not immediately recognise defeats the purpose — the bridging tracks are the point.
  3. Check the attribution when something surprises you. Finding out a track came from someone whose taste you thought you knew is half the value.
  4. Save what lands. Anything from the blend that you keep returning to should move into your own library, where it starts feeding your own recommendations.

If you want the live version of the same idea — same room, same second — that is a Jam. Blend is the version that works when nobody has to be free at the same time.

Frequently asked questions

What is a Blend on Melodius?

A shared playlist generated from the listening histories of everyone in it. It mixes music you already share, tracks that bridge the gap between your tastes, and each member's own favourites, and it refreshes as everyone keeps listening.

How is the taste match percentage calculated?

By comparing each member's listening as a weighted profile of artists and genres, with artist overlap counting for more than genre overlap. Sharing specific artists indicates far more in common than sharing a broad genre label.

How many people can be in a Blend?

A small group — up to ten members. With more people the shared core grows and each individual's section of the playlist gets smaller.

Why does my Blend look empty or thin?

Usually because one or more members do not have much listening history yet. A blend is computed from play history, so a new account contributes very little until it has listened to something.

Does a Blend update automatically?

Yes. It refreshes on a schedule as members keep listening, and what people are playing right now nudges it slightly within a bounded limit, so one heavy session cannot take the playlist over.

More on this topic

// Topic

Listening together

Synchronised sessions, shared queues and playlists built from two tastes.