Nodes/comfyui-llamacpp/llama.cpp ADV++ Prompt
ComfyUI Node

llama.cpp ADV++ Prompt

Runs multimodal generation with reusable prompt templates, token bans, and structured-output constraints.

By Setmaster·Created 7 months ago·Updated about a month ago· 4
llama.cpp ADV++ Prompt
  • trigger
  • token_ban
  • image_1
  • image_2
  • image_3
  • image_4
  • image_5
  • image_6
  • image_7
  • image_8
  • image_9
  • image_10
  • structured_output
  • connection
  • response
  • thinking
  • success
◄templateEmpty►
◄prompt►
◄image_amount2►
◄model(use running model)►
◄server_url►
◄system_prompt►
◄enable_thinkingtrue►
◄max_tokens2048►
◄temperature0.70►
◄top_p0.90►
◄top_k40►
◄min_p0.05►
◄repeat_penalty1.10►
◄presence_penalty0.0►
◄frequency_penalty0.0►
◄seed0►
◄keep_contextfalse►
◄enable_chainingfalse►
◄enable_token_bantrue►
◄stop_sequences►
◄api_key_envLLAMACPP_API_KEY►
◄verify_tlstrue►
◄request_timeout300►
◄include_image_batchfalse►
CategoryLlamaCpp

Inputs (38)

NameTypeDefaultDescription
templateCOMBOEmptyApply a bundled prompt template before generation.
promptSTRINGThe user prompt to send to the LLM
image_amountINT20–10Number of image input slots to show
modeloptCOMBO(use running model)Model for router mode, or the running direct model.
server_urloptSTRINGLeave empty to use the server owned by this node pack. Attached endpoints are never implicitly stopped.
system_promptoptSTRINGSystem prompt that defines model behavior.
enable_thinkingoptBOOLEANtrueRequest thinking/reasoning from compatible models.
max_tokensoptINT20481–131072Maximum number of tokens to generate.
temperatureoptFLOAT0.700–2Sampling randomness. Lower values are more deterministic.
top_poptFLOAT0.900–1Keep tokens within this cumulative probability mass.
top_koptINT400–200Sample from the top K tokens. 0 disables top-k filtering.
min_poptFLOAT0.050–1Discard tokens below this probability relative to the best token.
repeat_penaltyoptFLOAT1.101–2Penalize recently repeated tokens. 1.0 disables the penalty.
presence_penaltyoptFLOAT0.0-2–2Penalize tokens that have appeared at least once.
frequency_penaltyoptFLOAT0.0-2–2Penalize tokens in proportion to how often they appeared.
seedoptINT00–2147483647Random seed
keep_contextoptBOOLEANfalseReuse a matching prompt-prefix KV cache. This is not chat history.
enable_chainingoptBOOLEANfalseCompatibility toggle. A connected trigger already controls ordering.
triggeropt*Optional dependency input used to sequence execution.
token_banoptLOGIT_BIASToken ban list from a llama.cpp Token Ban node.
enable_token_banoptBOOLEANtrueEnable or disable the connected token ban list.
stop_sequencesoptSTRINGStop sequences. JSON arrays preserve commas and whitespace.
api_key_envoptSTRINGLLAMACPP_API_KEYEnvironment variable containing the API key. The secret is not serialized.
verify_tlsoptBOOLEANtrueVerify HTTPS certificates.
request_timeoutoptINT3001–86400Overall generation deadline in seconds.
include_image_batchoptBOOLEANfalseSend every image in each connected IMAGE batch.
image_1optIMAGEOptional image 1. Visibility follows image_amount.
image_2optIMAGEOptional image 2. Visibility follows image_amount.
image_3optIMAGEOptional image 3. Visibility follows image_amount.
image_4optIMAGEOptional image 4. Visibility follows image_amount.
image_5optIMAGEOptional image 5. Visibility follows image_amount.
image_6optIMAGEOptional image 6. Visibility follows image_amount.
image_7optIMAGEOptional image 7. Visibility follows image_amount.
image_8optIMAGEOptional image 8. Visibility follows image_amount.
image_9optIMAGEOptional image 9. Visibility follows image_amount.
image_10optIMAGEOptional image 10. Visibility follows image_amount.
structured_outputoptSTRUCTURED_OUTPUTJSON schema, JSON object, or GBNF constraint.
connectionoptLLAMACPP_CONNECTIONOptional reusable local or remote connection profile.

Outputs (3)

NameTypeDescription
responseSTRINGGenerated multimodal or structured response text.
thinkingSTRINGReasoning content reported separately by compatible models.
successBOOLEANWhether generation completed successfully.