2 months ago
bd90d76
Davy* feat(separation): default new separations to four stems Implements the corrected #182 decision: the model natively outputs four stems and the two-stem projection costs the same inference time, so four-stem becomes the default for NEW separations. The stem-mode setting keeps its existing meaning (processing output for new songs) — there is no playback-view concept: playing a four-stem song while the two-stem mode is selected simply plays all four stems, which sounds identical until per-stem volumes are touched. - StemMode::default() flips to FourStem (config, settings store, mocks); existing separations and explicit user choices are unaffected. - Space-sensitive users keep the OPTIONAL compress-to-two-stem paths (per-song downgrade and the library-wide maintenance action), and can still select two-stem output for new processing. - The release smoke pins TwoStem explicitly: its validator asserts the vocals + accompaniment artifact pair, which four-stem does not write. - Settings copy updated in en/zh-CN to explain the default and the space trade-off. Closes #182. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * test(e2e): keep the mock session on the two-stem slider layout The Playwright geometry specs assert the two-stem stem-button layout; the mock is an e2e fixture, not a carrier of the product default. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Parentf6b4e73