Nodes/ComfyUI-FL-MiniMaxH3/FL MiniMax H3 LoRA Block Loader
ComfyUI Node

FL MiniMax H3 LoRA Block Loader

Load one H3 LoRA with per-block strengths. Only this adapter is scaled; existing model patches are preserved. All multipliers at 1 match normal model-only LoRA loading.

By filliptm·Created about a month ago·Updated about 14 hours ago· 17
FL MiniMax H3 LoRA Block Loader
  • model
  • model
  • block_report
lora_name
strength_model1.00
blocks_strength1.00
refiner_strength1.00
other_strength1.00
block_overrides
CategoryFL/MiniMax H3/Model

Inputs (7)

NameTypeDefaultDescription
modelMODEL
lora_nameCOMBO0 options:
strength_modelFLOAT1.00-10–10
blocks_strengthFLOAT1.00-10–10Multiplier for main transformer blocks.
refiner_strengthFLOAT1.00-10–10Multiplier for token-refiner blocks; not the external Qwen encoder.
other_strengthFLOAT1.00-10–10Multiplier for adapter targets outside both stacks: projections, time embedding and final layer.
block_overridesSTRINGOne rule per line: blocks.0-9=0.5 or refiner.1=0. Zero-based, inclusive ranges. Replaces the group multiplier; overall strength still applies. Later rules win. # comments allowed.

Outputs (2)

NameTypeDescription
modelMODEL
block_reportSTRING