One way to look at OP_RETURN usage by "L2" projects, is to get a sense of what future op codes could look like. If people keep sticking zero knowledge proofs there, maybe at some it makes sense to just add a zero knowledge proof output type. Obviously that's a _very_ long time out, and there might be very good reasons to never do it. But you can look at it as a (permissionless) draft.
80 bytes is enough for a pointer plus metadata — the most efficient form of layer 2 scaling.
The principle is simple: instead of putting the entire image (or data) on-chain, you embed a pointer or link and store the actual content off-chain — via IPFS, Nostr, Blossom, Nosdav, CIDs, or future solutions built for this purpose.
This approach is cheaper, scalable, and keeps Bitcoin lean and mean for financial transactions.
Ossification is good — aside from essential maintenance updates like cleanup or adjusting mempool limits to match modern hardware.
