Nodes/WAS Node Suite v3/Pika H3 Flex LoRA
ComfyUI Node Runs on cloud

Pika H3 Flex LoRA

Turn a MiniMax H3 fl2va or ref2va model into Pika H3 Flex: the matching Pika adapter, the release's video and audio shifts of 12 and 3, and its fp32 AdaLN. Sample it on Pika H3 Flex Sampling's sigmas.

By WASasquatch·Created 4 years ago·Updated 3 days ago· 1,900
Pika H3 Flex LoRA
  • model
  • model
◄lora_nameNone►
◄strength1.00►
◄modebypass►
◄adaln_fp32true►
CategoryWAS Suite/Sampling

Inputs (5)

NameTypeDefaultDescription
modelMODELMiniMax H3 from Load Diffusion Model: fl2va for the fl2va adapter, ref2va for the ref2va adapter.
lora_nameCOMBONoneAdapter in models/loras: `pika_h3_flex_fl2va` for text and frames, `pika_h3_flex_ref2va` for references; `None` = shifts and AdaLN only. A wired name, or a full path inside that model's folders such as `E:\models\loras\style.safetensors`, works too.
strengthFLOAT1.00-2–2Adapter scale: `1.0` = the release; `0` = the base model.
modeCOMBObypass`bypass` = adapter computed beside each layer, base weights untouched, exact on int8 and w6a8 models; `merge` = adapter added into the weights.
adaln_fp32BOOLEANtrue`true` = AdaLN SiLU on the fp32 time embedding, as the release; `false` = on the embedding in the compute dtype, as core. Pruned models run fp32.

Outputs (1)

NameTypeDescription
modelMODELPika H3 Flex at video shift 12 and audio shift 3.