ComfyUI_LLM_SDXL_Adapter_Additions

Additional nodes for use alongside ComfyUI_LLM_SDXL_Adapter with enhanced T5-GEMMA text encoding and improved adapter loading capabilities.

By lRemixl·Created 5 months ago·Updated 4 months ago· 0

lRemixl/ComfyUI_LLM_SDXL_Adapter_Additions

Nodes—

On cloudLocal install

Stars0

Updated4 months ago

Readme

ComfyUI_LLM_SDXL_Adapter_Additions

Additional nodes for use alongside ComfyUI_LLM_SDXL_Adapter

t5gemma_text_encode_v2.py

Node: T5GEMMATextEncoder++

Purpose: Modified from the devlop branch of ComfyUI_LLM_SDXL_Adapter. Adds chunking for prompts greater than 512 tokens. Also significantly simplifies workflow.

llm_adapter_loader_explicit.py

Node: LLMAdapterLoaderUnfused

Purpose: Load T5 Gemma adapter that has unfused QKV. Compared to the regular version that has them fused. Allows for slightly better quality when loading adapters that have been trained with unfused QKVs.