Yes, and without a soft fork. In Gibson's https://github.com/AdamISZ/ln-gap/blob/master/docs/paper/lngap-short.pdf, LN-GAP covers two-party contracts: the channel commitment carries an extra contract output with a pre-signed dispute graph, moves are exchanged off-chain, and a dispute settles in at most 4 on-chain transactions however long the game runs.

The catch is the "venue", a committee that only attests whether a move was published before its deadline. It never holds funds, but as https://bitcoinops.org/en/newsletters/2026/10/16/#offchain-contracts-in-lightning-channels notes, if more than half its members collude with one player, funds can be stolen.

Sizes measured in the paper with a 5-member venue: a stall claim costs the claimant two small transactions (188 vB plus 193-201 vB), but a chess rebuttal is about 24,121 vB, paid by the side answering the claim, with in-channel deposits meant to make the offender carry those costs. The https://github.com/AdamISZ/ln-gap says the code is AI-generated, a proof of concept run on regtest, and not fit for real money.