Issues

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

  huggingface/diffusers#14970 · opened on Oct 6, 2026 · Release 0.42.0
  2

• get_active_adapters() returns empty for active text-encoder-only LoRA adapters

  huggingface/diffusers#14955 · opened on Oct 6, 2026 · Release 0.42.0

• PNDMScheduler.set_timesteps crashes with NumPy broadcasting error for fewer than 4 inference steps

  huggingface/diffusers#14953 · opened on Oct 6, 2026

• ModelMixin.from_pretrained on a sharded checkpoint fails under HF_HUB_OFFLINE=1 even when fully cached (_get_checkpoint_shard_files still calls model_info)

  huggingface/diffusers#14952 · opened on Oct 5, 2026
  1

• Propose translation to romanian for the diffusers hugging face course

  huggingface/diffusers#14939 · opened on Oct 4, 2026

• Qwen-Image LoRAs in lora_A/lora_B format lose their .alpha and load at the wrong strength

  huggingface/diffusers#14938 · opened on Oct 4, 2026

• Qwen-Image-2.1 LoRAs with a fused img_mlp.gate_up load partially without an error

  huggingface/diffusers#14937 · opened on Oct 4, 2026

• QwenImage21: model.to(bf16) (as in train_dreambooth_lora_qwenimage21.py) rounds the timestep-embedding freqs buffer, so LoRA training sees a different embedding than inference

  huggingface/diffusers#14935 · opened on Oct 4, 2026

• StableAudio3DiTModel: loading with torch_dtype=bf16/fp16 collapses RoPE positions (arange built in inv_freq dtype)

  huggingface/diffusers#14934 · opened on Oct 4, 2026

• from_single_file() support for AnimaTextConditioner

  huggingface/diffusers#14930 · opened on Oct 3, 2026
  2

• CosmosTransformer3DModel.from_single_file() does not work with the model.diffusion_model. prefix

  huggingface/diffusers#14929 · opened on Oct 3, 2026

• Qwen-Image-2.1 VAE tiling: RGB/alpha seams and a full-context decode workaround on ROCm

  huggingface/diffusers#14928 · opened on Oct 3, 2026

• Windows: from_single_file original_config and fp16_safetensors crash with UnicodeDecodeError under non-UTF-8 locales

  huggingface/diffusers#14927 · opened on Oct 3, 2026

• Add a Hugging Face Jobs guide to the docs

  huggingface/diffusers#14924 · opened on Oct 2, 2026

• Enable Taylorseer for all Flux Pipelines

  huggingface/diffusers#14922 · opened on Oct 1, 2026

• fal FLUX.1 Kontext LoRAs with unprefixed embedder keys fail in _convert_fal_kontext_lora_to_diffusers

  huggingface/diffusers#14917 · opened on Sep 30, 2026

• CFG cutoff callbacks keep a single embedding row, so they break for num_images_per_prompt > 1 or several prompts

  huggingface/diffusers#14907 · opened on Sep 29, 2026
  1

• Required weight pass on a singlefile

  huggingface/diffusers#14903 · opened on Sep 29, 2026

• numpy_to_pil removes singleton spatial dimensions from grayscale images

  huggingface/diffusers#14901 · opened on Sep 29, 2026
  1

• Legacy final_offload_hook checks are dead in 14 pipelines; KandinskyPriorPipeline skips maybe_free_model_hooks() when negative_prompt is set

  huggingface/diffusers#14896 · opened on Sep 28, 2026
  1

• Cosmos2VideoToWorldPipeline conditions on the entire input video — no num_conditional_frames, so a full clip is reconstructed instead of continued

  huggingface/diffusers#14892 · opened on Sep 28, 2026
  2

• Converted sigma schedules with final_sigmas_type="sigma_min" repeat the last sigma, and the third-order and heun updates of DPMSolverMultistepScheduler and UniPCMultistepScheduler return NaN on that zero-width final step

  huggingface/diffusers#14890 · opened on Sep 28, 2026

• QwenImage21 FlexAttention mask fails to compile on MPS because of boolean ~

  huggingface/diffusers#14889 · opened on Sep 28, 2026

• UniPCMultistepScheduler fails in torch.linalg.solve under a float64 default dtype: the unit entry of rks takes the default dtype

  huggingface/diffusers#14888 · opened on Sep 27, 2026

• UniPCMultistepScheduler returns nan with solver_type="bh1", and inf or nan with lower_order_final=False or predict_x0=False, on the final step to sigma = 0

  huggingface/diffusers#14887 · opened on Sep 27, 2026