24h GitHub PR Update
24h pull request roundup with 13 new PRs across 1 active repos in the tracked set.
https://github.com/Comfy-Org/ComfyUI/tree/master
- #13070 Revert “fix: run text encoders on MPS GPU instead of CPU for Apple Silicon” by @comfyanonymous: Rolls back the Apple Silicon text-encoder GPU change to stabilize behavior pending a safer implementation.
- #13068 fix: make Content-Disposition header RFC 2183 compliant (Fixes #8914) by @junagent: Brings download headers into RFC compliance for more reliable client/browser file handling.
- #13067 memory: Add more exclusion criteria to pinned read (fixes corrupt outputs - rare cases) by @rattus128: Adds stricter pinned-read guards to prevent rare corruption scenarios in memory-sensitive runs.
- #13062 ltx: vae: implement chunked encoder + CPU IO chunking (Big VRAM reductions) by @rattus128: Introduces chunked LTX VAE encode/decode paths that materially reduce VRAM pressure.
- #13060 [API Nodes] mark seedream-3-0-t2i and seedance-1-0-lite models as deprecated by @bigcat88: Marks older API-node model entries deprecated to steer users toward current supported options.
- #13059 feat: add get_execution_environment() API by @deepme987: Adds an API for runtime environment introspection, useful for adaptive node behavior.
- #13058 Main by @kristentr: Proposes branch sync/update changes under a generic mainline PR.
- #13057 Main (#3) by @kristentr: Submits an additional mainline sync/update iteration.
- #13056 fp16 intermediates doen’t work for some text enc models. by @comfyanonymous: Fixes fp16 intermediate dtype handling for text-encoder models that failed in mixed-precision.
- #13054 fix: make Content-Disposition header RFC 2183 compliant by @junagent: Corrects attachment header formatting for standards-compliant file download behavior.
- #13053 Fix regression. by @comfyanonymous: Patches a newly introduced regression impacting recent workflows.
- #13052 Further Reduce LTX VAE decode peak RAM usage by @kijai: Further lowers LTX VAE decode memory spikes to improve run stability on lower-VRAM systems.
- #13051 Make more intermediate values follow the intermediate dtype. by @comfyanonymous: Aligns more intermediate operations with configured dtype to reduce conversion errors and inconsistencies.
https://github.com/shootthesound/comfyUI-Realtime-Lora
- no new pull requests in last 24h
https://github.com/ostris/ai-toolkit
- no new pull requests in last 24h
https://github.com/AkaneTendo25/musubi-tuner/tree/ltx-23
- no new pull requests in last 24h
https://github.com/ArtDesignAwesome/ai-toolkit_BIG-DADDY-VERSION
- no new pull requests in last 24h
https://github.com/kijai/ComfyUI-KJNodes
- no new pull requests in last 24h
https://github.com/Lightricks/LTX-Desktop
- no new pull requests in last 24h
https://github.com/Lightricks/ComfyUI-LTXVideo
- no new pull requests in last 24h
https://github.com/Lightricks/LTX-2
- no new pull requests in last 24h
https://github.com/wildminder/awesome-ltx2
- no new pull requests in last 24h