New Model Request Reply

Workflow runs for New Model Request Reply

• LTX2 HDR reference preprocessing applies an unintended stride-aligned resize

  skipped:
  New Model Request Reply#168:Issue#15037openedbyShenrui-Ma
  Oct 14, 20261s
  More actions for workflow run 168

• Qwen Image 2.1 with torch.compile triggers a recompile on each run with varying image size

  skipped:
  New Model Request Reply#167:Issue#15027openedbygeekuillaume
  Oct 14, 20262s
  More actions for workflow run 167

• SDXL invisible watermark: apply_watermark encodes float input, and about a third of outputs are not detected by Stability's detect.py

  skipped:
  New Model Request Reply#166:Issue#15025openedbyCyberNativeAI
  Oct 14, 20261s
  More actions for workflow run 166

• Modular pipelines: warn when a denoiser_input_fields entry is not passed to the transformer

  skipped:
  New Model Request Reply#165:Issue#15022openedbyyiyixuxu
  Oct 13, 20261s
  More actions for workflow run 165

• add_noise() before set_timesteps() uses the wrong noise level (DPMSolverMultistep, DPMSolverSinglestep, UniPC, DEIS; also SanaPipeline)

  skipped:
  New Model Request Reply#164:Issue#15021openedbyFilipLangiewicz
  Oct 13, 20267s
  More actions for workflow run 164

• [LoRA] Text-encoder LoRA loading assumes CLIP module names and fails for T5 / SmolLM3 / ChatGLM text encoders

  skipped:
  New Model Request Reply#163:Issue#15020openedbySergasgr
  Oct 12, 202611s
  More actions for workflow run 163

• offload_state_dict=True leaves CPU model weights on meta after loading

  skipped:
  New Model Request Reply#162:Issue#15018openedbyShenrui-Ma
  Oct 12, 2026Skipped
  More actions for workflow run 162

• gfx1201 (RDNA4) SageAttention attention backend: aiter package or Hub kernel?

  skipped:
  New Model Request Reply#161:Issue#15014openedbyShan2L
  Oct 12, 20261s
  More actions for workflow run 161

• VaeImageProcessor.denormalize() raises AttributeError for NumPy inputs

  skipped:
  New Model Request Reply#160:Issue#15011openedbySrijanRoy123-github
  Oct 12, 20261s
  More actions for workflow run 160

• [modular] SequentialPipelineBlocks gets required wrong in both inputs and required_inputs

  skipped:
  New Model Request Reply#159:Issue#15009openedbyyiyixuxu
  Oct 11, 20261s
  More actions for workflow run 159

• [Z-Image] joint_attention_kwargs={"scale": ...} is silently ignored when loading LoRA weights

  skipped:
  New Model Request Reply#158:Issue#15007openedbySergasgr
  Oct 11, 20261s
  More actions for workflow run 158

• WanVideoToVideoPipeline: "No cache context is set" when caching is enabled

  skipped:
  New Model Request Reply#157:Issue#15006openedbyfusheng-ji
  Oct 11, 20261s
  More actions for workflow run 157

• Add TripoSplat

  completed successfully:
  New Model Request Reply#156:Issue#15005openedbyhlky
  Oct 11, 20269s
  More actions for workflow run 156

• KDPM2DiscreteScheduler produces NaN on final step due to zero-sigma interpolation

  skipped:
  New Model Request Reply#155:Issue#15002openedbyriparanta
  Oct 10, 20267s
  More actions for workflow run 155

• load_video() leaks temporary files when remote video decoding fails

  skipped:
  New Model Request Reply#154:Issue#14994openedbySrijanRoy123-github
  Oct 9, 20261s
  More actions for workflow run 154

• WanCausalConv3d copies every activation with F.pad. Let the conv do the spatial padding.

  skipped:
  New Model Request Reply#153:Issue#14993openedbywill-rice
  Oct 8, 20261s
  More actions for workflow run 153

• Add an optional Transformer Engine FP8 training example for FLUX

  skipped:
  New Model Request Reply#152:Issue#14992openedbysbhavani
  Oct 8, 20261s
  More actions for workflow run 152

• MiniMax-H3: _decode_video_file turns ±90° display-rotated videos upside down

  skipped:
  New Model Request Reply#151:Issue#14985openedbyhsbcelebi
  Oct 7, 20262s
  More actions for workflow run 151

• LTX-2.5 SDR-To-HDR IC-LoRA: how should it be structured?

  skipped:
  New Model Request Reply#150:Issue#14981openedbychristopher5106
  Oct 7, 20261s
  More actions for workflow run 150

• LTX-2: decouple the RoPE time base from the playback frame rate (Slow-Motion-Control LoRA)

  skipped:
  New Model Request Reply#149:Issue#14980openedbychristopher5106
  Oct 7, 20261s
  More actions for workflow run 149

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

  skipped:
  New Model Request Reply#148:Issue#14970openedbyandre-menutole
  Oct 6, 20262s
  More actions for workflow run 148

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

  skipped:
  New Model Request Reply#147:Issue#14955openedbySrijanRoy123-github
  Oct 6, 20261s
  More actions for workflow run 147

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

  skipped:
  New Model Request Reply#146:Issue#14953openedbySrijanRoy123-github
  Oct 6, 20262s
  More actions for workflow run 146

• 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)

  skipped:
  New Model Request Reply#145:Issue#14952openedbyquinncheong
  Oct 5, 20261s
  More actions for workflow run 145

• Propose translation to romanian for the diffusers hugging face course

  skipped:
  New Model Request Reply#144:Issue#14939openedbypop123-ux
  Oct 4, 20261s
  More actions for workflow run 144