24h GitHub Repo Summary

GitHub Tracked repos · Mar 19, 2026, 10:03 p.m.

24h commit roundup with 14 commits across 2 active repos in the tracked set.

8458ae2: Reverts the earlier Apple Silicon text-encoder GPU change to restore stable behavior while that fix is reworked.
fd0261d: Reduces tiled decode peak memory to lower OOM risk on constrained GPUs.
ab14541: Tightens pinned-read exclusion rules to avoid rare memory-path corruption edge cases.
6589562: Adds chunked LTX VAE encoding and CPU I/O chunking for large practical VRAM savings.
fabed69: Adds chunked LTX VAE encoding and CPU I/O chunking for large practical VRAM savings.
f6b869d: Fixes fp16 intermediate handling for text-encoder models that were failing under reduced precision.
56ff88f: Fixes a regression introduced in recent changes so affected workflows run correctly again.
9fff091: Further lowers LTX VAE decode RAM peaks to improve reliability for longer or heavier generations.
dcd6595: Makes intermediate tensor operations adhere to selected dtype more consistently, reducing mismatch issues.

d5c9b0a: Improves Set/Get separation between workflow loading and node copying to prevent cross-context confusion.
535d7cb: Hardens LTXVAudioVideoMask validation so invalid masks fail more gracefully.
297a245: Fixes Set/Get context-menu insertion when working inside subgraphs.
6dfca48: Updates node help popup support to the newer 2.0 behavior.
00af95b: Delivers a full Set/Get subsystem overhaul for cleaner node-linking workflows.