Nodes/LLM Text Processor/LLM Text Processor
ComfyUI Node

LLM Text Processor

A ComfyUI node in LLM Text Processor with 19 inputs and 3 outputs.

By KingManiya·Created 4 months ago·Updated 4 months ago· 21
LLM Text Processor
  • image
  • RESPONSE
  • REASONING
  • PERF
model
mmprojnone
system_prompt
promptDescribe this image in detail.
max_tokens2048
temperature0.70
top_p0.80
top_k20
repeat_penalty1.00
ctx_size8192
memory_modeauto
n_gpu_layers99
n_cpu_moe_layers1
seed1
timeout_seconds300
reasoningoff
enable_processingtrue
extra_args
CategoryLLM Text Processor

Inputs (19)

NameTypeDefaultDescription
modelCOMBOGGUF model loaded from ComfyUI/models/LLM. mmproj files are hidden from this list.
mmprojCOMBOnoneVision projector GGUF. Required when one or more images are connected.
system_promptCOMBOSystem prompt preset from ComfyUI/models/LLM/prompts, or none.
promptSTRINGDescribe this image in detail.User prompt sent to the selected model.
max_tokensINT20481–32768Maximum number of tokens to generate.
temperatureFLOAT0.700–2Sampling temperature. Lower is more deterministic.
top_pFLOAT0.800–1Nucleus sampling threshold.
top_kINT201–1000Top-K sampling cutoff.
repeat_penaltyFLOAT1.000–3Penalty applied to repeated tokens.
ctx_sizeINT8192512–1048576Context window size in tokens. Use a value supported by the selected GGUF; larger context uses more VRAM.
memory_modeCOMBOautoAdvanced memory placement mode: auto, gpu_layers, cpu_moe_layers, or gpu_and_cpu_moe_layers.
n_gpu_layersINT99-1–999Used only in gpu_layers and gpu_and_cpu_moe_layers modes. Number of model layers to place on the GPU.
n_cpu_moe_layersINT11–999Used only in cpu_moe_layers and gpu_and_cpu_moe_layers modes. Number of MoE layers to keep on the CPU.
seedINT1-1–4294967295Random seed. Use -1 for a random seed.
timeout_secondsINT30010–3600Maximum time to wait before generation is stopped.
reasoningCOMBOoffReasoning output mode.
imageoptIMAGEOptional image input. A single image or ComfyUI batch is passed to llama.cpp.
enable_processingoptBOOLEANtrueWhen enabled, run normal node processing. When disabled, forward the input prompt directly as RESPONSE.
extra_argsoptSTRINGOptional advanced llama.cpp parameters. Leave empty for normal use.

Outputs (3)

NameTypeDescription
RESPONSESTRINGFinal model response with reasoning blocks removed.
REASONINGSTRINGExtracted reasoning when present in model output.
PERFSTRINGllama.cpp prompt and generation speed.