Release 0.42.0

Issues

• [modular] improve auto offload

  documentationmodular-pipelinessize/Ltestsutils
  Unread
  huggingface/diffusers#14304·yiyixuxu opened Jul 27, 2026 · Updated Oct 2, 2026
  9

• [modular] LTX-2.5: two-stage generation as one pipeline

  documentationmodular-pipelinessize/Ltestsutils
  Unread
  huggingface/diffusers#14612·yiyixuxu opened Aug 27, 2026 · Updated Oct 2, 2026
  8

• [Modular]improve loop block

  documentationmodular-pipelinessize/Ltestsutils
  Unread
  huggingface/diffusers#14159·yiyixuxu opened Jul 10, 2026 · Updated Oct 6, 2026
  33

• Allow tensor parallelism and context parallelism in one ParallelConfig

  hooksmodelsno-issue-neededsize/Ltests
  Unread
  huggingface/diffusers#14725·whn09 opened Sep 7, 2026 · Updated Oct 14, 2026
  9

  

• fix: scope save_pretrained shard cleanup to the saved variant (avoid deleting another variant's shards)

  fixes-issuemodelssize/Mtests
  Unread
  huggingface/diffusers#14844·AloysJehwin opened Sep 22, 2026 · Updated Oct 4, 2026
  1

• feat: implement cache_context for pab and fastercache.

  documentationhooksmodelsmodular-pipelinespipelinessize/Ltests
  Unread
  huggingface/diffusers#14900·sayakpaul opened Sep 29, 2026 · Updated Oct 14, 2026
  16

• feat: cp support in qwenimage 2.1

  modelssize/Ltests
  Unread
  huggingface/diffusers#14817·sayakpaul opened Sep 20, 2026 · Updated Oct 7, 2026·Draft
  4

• Per-module alphas of kohya text encoder LoRAs collapse to the most common alpha (alpha_pattern keys stop at the parent module)

  buglorapipelines
  huggingface/diffusers#14970 · Filter by author andre-menutole opened on Oct 6, 2026 · Release 0.42.0
  7

• fix: derive text-encoder alpha_pattern keys from the LoRA module name

  fixes-issuesize/Mtestsutils
  Unread
  huggingface/diffusers#14972·akhil9tiet opened Oct 6, 2026 · Updated Oct 13, 2026
  8

• fix: prefer latest flash attn2

  modelssize/S
  Unread
  huggingface/diffusers#15000·drbh opened Oct 9, 2026 · Updated Oct 14, 2026
  4