Nodes/Deno Custom Nodes/(Deno) Local LLM Loader
ComfyUI Node

(Deno) Local LLM Loader

DENO Custom Nodes vunknown. The DENO info button checks Comfy Registry and marks this node when an update is available. Call a local Ollama, LM Studio, llama.cpp, vLLM, or Custom model from ComfyUI and help rewrite or review prompt text. An optional IMAGE input can be attached to the local model call. Use a vision-capable local model for image review. Designed for prompt-batcher workflows: use the in-node Prompt field or connect STRING into Prompt, and this node processes the whole prompt batch in one execution so the local LLM can stay loaded until the batch is complete. Only localhost / 127.0.0.1 servers are allowed by default. Advanced users can allow a specific LAN LLM server with DENO_LOCAL_LLM_ALLOWED_HOSTS=LAN-IP:port on the ComfyUI host; this opt-in is not stored in workflows.

By Deno2026·Created 4 months ago·Updated 24 days ago· 116
(Deno) Local LLM Loader
  • image
  • result
◄providerOllama►
◄ollama_model►
◄lm_studio_model►
◄custom_server_urlhttp://127.0.0.1:8000/v1►
◄custom_model►
◄system_prompt►
◄thinkingfalse►
◄seed1►
◄seed_modefixed►
◄model_memoryUnload after run►
◄keep_minutes5►
◄comfy_vram_policyAuto: unload only before first LLM call►
◄prompt►
CategoryDeno/LLM

Inputs (14)

NameTypeDefaultDescription
providerCOMBOOllamaChoose the local provider to call: Ollama, LM Studio, llama.cpp, vLLM, or Custom.
ollama_modelCOMBOOllama model name detected from your local Ollama server.
lm_studio_modelCOMBOLM Studio model name detected from your local LM Studio server.
custom_server_urlSTRINGhttp://127.0.0.1:8000/v1Local OpenAI-compatible server URL used by llama.cpp, vLLM, or Custom.
custom_modelSTRINGModel name to call on the selected local OpenAI-compatible server.
system_promptSTRINGOptional instructions that steer how the local LLM rewrites or reviews the prompt.
thinkingBOOLEANfalseAsk for model thinking/reasoning when the selected local model supports it.
seedINT10–4294967295Seed used for local LLM generation.
seed_modeCOMBOfixedHow the visible seed changes after each queued run.
model_memoryCOMBOUnload after runChoose whether the local model unloads after the run or stays warm.
keep_minutesINT51–240How long to keep the local model loaded when Keep loaded is selected.
comfy_vram_policyCOMBOAuto: unload only before first LLM callChoose whether ComfyUI models should be unloaded before local LLM work.
promptSTRINGMain prompt text. The in-node textarea and STRING socket feed this same backend input.
imageoptIMAGEOptional image sent to a vision-capable local model.

Outputs (1)

NameTypeDescription
resultSTRINGLocal LLM answer or extracted final prompt. Batched prompts return a STRING list.