Issues

• [bug]: Qwen-Image LoRAs with diffusion_model. or base_model.model. key prefixes are not identified and apply to no layer

  bug
  invoke-ai/InvokeAI#10031 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: FLUX kohya and OneTrainer LoRAs with layers outside the transformer blocks are not identified or converted

  bug
  invoke-ai/InvokeAI#10030 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: Anima kohya LoRAs: layers outside the DiT blocks are not applied

  bug
  invoke-ai/InvokeAI#10028 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: Anima LoRAs with diffusers module names are identified as Qwen-Image

  bug
  invoke-ai/InvokeAI#10027 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: FLUX LoRAs in diffusion_model. + lora_A/lora_B layout with .alpha keys fail to load

  bug
  invoke-ai/InvokeAI#10025 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: FLUX.1 diffusers-format LoRAs with text-encoder layers fail to load

  bug
  invoke-ai/InvokeAI#10024 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: FLUX.1 diffusers LoRA converter maps both guidance-embedder layers onto guidance_in

  bug
  invoke-ai/InvokeAI#10023 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: FLUX final-layer modulation LoRA layer is not converted (FLUX.1 diffusers: load fails, FLUX.2 BFL: layer skipped)

  bug
  invoke-ai/InvokeAI#10022 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: Qwen-Image text-encoder LoRA layers are dropped (silently for kohya keys)

  bug
  invoke-ai/InvokeAI#10021 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: Qwen-Image LoKr with dotted keys fails to load; LoHa and factorized-w1 LoKr are not identified

  bug
  invoke-ai/InvokeAI#10020 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: Z-Image, Qwen-Image and Anima LoRA converters ignore .alpha and drop DoRA magnitudes on lora_A/lora_B layers

  bug
  invoke-ai/InvokeAI#10019 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: LoKr and LoHa layers with a DoRA magnitude fail to load (KeyError: 'lora_up.weight')

  bug
  invoke-ai/InvokeAI#10018 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: LoHa on Conv2d layers with conv-shaped factors fails at generation time

  bug
  invoke-ai/InvokeAI#10017 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: DoRA dora_scale stored on the output axis (LyCORIS default) is applied on the input axis

  bug
  invoke-ai/InvokeAI#10016 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: LyCORIS LoKr with a factorized w2 is misidentified (SD1.x/SD2.x install as Unknown, FLUX.2 as FLUX.1)

  bug
  invoke-ai/InvokeAI#10015 · Filter by author Pfannkuchensack opened on Oct 25, 2026

• [bug]: VAE encode/decode paths move inputs to the session device instead of the VAE's compute device, breaking cpu_only VAEs on GPU machines

  invoke-ai/InvokeAI#10014 · Filter by author Pfannkuchensack opened on Oct 24, 2026

• [bug]: Working-memory calibration plan ignores a model's default VAE

  invoke-ai/InvokeAI#10012 · Filter by author Pfannkuchensack opened on Oct 24, 2026

• [bug]: Working-memory calibration plan reports FLUX.1, Z-Image and SD3 models without their encoders as supported

  invoke-ai/InvokeAI#10011 · Filter by author Pfannkuchensack opened on Oct 24, 2026

• [bug]: Generate picks an SDNQ FLUX.2 pipeline as FLUX.2 [dev] component source, which the [dev] loader rejects

  invoke-ai/InvokeAI#10010 · Filter by author Pfannkuchensack opened on Oct 24, 2026

• [bug]: Model cache offloads without MODEL_LOAD_LOCK can strand weights on meta during a concurrent construction

  invoke-ai/InvokeAI#10009 · Filter by author Pfannkuchensack opened on Oct 24, 2026

• [bug]: ERNIE-Image Heun scheduler refuses every partial denoise (img2img, inpaint, Canvas)

  bug
  invoke-ai/InvokeAI#10003 · Filter by author Pfannkuchensack opened on Oct 23, 2026

• [bug]: [v7] Idle invokeai-web pins one CPU core at 100% (7.0.0a2-rocm, Radeon AI PRO R9700)

  bug
  invoke-ai/InvokeAI#9998 · Filter by author CarstenD74 opened on Oct 23, 2026
  5

• [bug]: SDNQ pipelines load their VAE in fp32, and a standalone SDNQ FLUX VAE in fp16 under auto

  invoke-ai/InvokeAI#9996 · Filter by author Pfannkuchensack opened on Oct 23, 2026

• [bug]: FLUX VAE nodes cast inputs to an SDNQ VAE's packed storage dtype instead of its compute dtype

  invoke-ai/InvokeAI#9995 · Filter by author Pfannkuchensack opened on Oct 23, 2026

• [enhancement]: Support moss image

  enhancement
  invoke-ai/InvokeAI#9992 · Filter by author Leroy-X opened on Oct 22, 2026
  1