ComfyUI Node

LlamaCPP Chat

Chat with llama.cpp. Use media_mode to select none / image / video / audio input. Set visualization=html to force HTML output and render it to html_image. Wire a LlamaCPP Visualizer HTML node to viz_settings to control render dimensions (optional, defaults to 800x600).

By pointave·Created 5 months ago·Updated 2 months ago· 2
LlamaCPP Chat
  • connectivity
  • options
  • images
  • video
  • audio
  • viz_settings
  • result
  • thinking
  • html_image
systemYou are an AI assistant.
promptHello!
thinkfalse
format
reset_sessiontrue
media_modenone
visualizationdisabled
CategoryLlamaCPP API

Inputs (13)

NameTypeDefaultDescription
systemSTRINGYou are an AI assistant.
promptSTRINGHello!
thinkBOOLEANfalse
formatCOMBO2 options: text, json
reset_sessionBOOLEANtrue
media_modeCOMBOnoneWhich optional media socket to process. none=text only (inputs ignored even if wired), image=still frames, video=temporal batch, audio=AUDIO input (Gemma4 E2B/E4B only)
visualizationCOMBOdisabledhtml: forces full HTML document output and renders the result to html_image via Playwright.
connectivityoptLLAMACPP_CONNECTIVITY
optionsoptLLAMACPP_OPTIONS
imagesoptIMAGEActive when media_mode=image
videooptIMAGEActive when media_mode=video. Batch treated as temporal frames. Max 60 frames.
audiooptAUDIOActive when media_mode=audio. Gemma4 E2B/E4B only. Max 30 s.
viz_settingsoptLLAMACPP_VIZ_SETTINGSWire a LlamaCPP Visualizer HTML node to control render width/height/delay. Optional — defaults to 800x600.

Outputs (3)

NameTypeDescription
resultSTRING
thinkingSTRING
html_imageIMAGE