ComfyUI Node
MiniMax H3 Power LoRA Stack
Stacked multi-LoRA loader for MiniMax H3 across every weight format. Handles the three things the stock loader and rgthree's Power Lora Loader get wrong on H3: * quantized bases -- w4a8/int4 weights are patched with an exact runtime branch instead of a lossy dequantize/requantize merge; * adaLN basis mismatch -- LoRAs trained against a dense checkpoint are rebased onto a pruned checkpoint's curve, instead of being skipped; * key conventions -- ai-toolkit, kohya, lycoris, peft and bare-prefix layouts all resolve against the model's own key set.
MiniMax H3 Power LoRA Stack
- model
- adaln_modality
- schedule
- MODEL
- report
◄quantized_layersauto►
◄adaln_portauto►
CategoryMiniMax-H3/lora
Inputs (5)
| Name | Type | Default | Description |
|---|---|---|---|
| modelopt | MODEL | — | |
| quantized_layersopt | COMBO | auto | auto: runtime branch for quantized layers, merge for the rest. branch: never modify a weight. merge: stock behaviour (destroys LoRAs on w4a8/int4). |
| adaln_portopt | COMBO | auto | Rebase adaLN LoRA pairs between dense (2688) and curve (8) checkpoints. Needs h3_silu_temb_grid.safetensors. |
| adaln_modalityopt | H3_MODALITY | Optional. Wire a MiniMax H3 adaLN Modality node here to scale every stacked LoRA's adaLN modulation per modality. | |
| scheduleopt | H3_SCHEDULE | Optional. Wire an H3 LoRA Schedule chain here to vary selected row strengths over the denoising trajectory. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | Patched model |
| report | STRING | Per-LoRA account of what was applied |