Nodes/ComfyUI-Prompt-Manager/Prompt Generator Options
ComfyUI Node

Prompt Generator Options

Configure LLM model, sampling parameters, and system prompt for the Prompt Generator.

By FranckyB·Created 10 months ago·Updated 23 days ago· 125
Prompt Generator Options
  • image2
  • image3
  • image4
  • image5
  • options
modelQwen3.5-9B-UD-Q4_K_XL.gguf
system_prompt_modeappend
system_prompt
use_model_default_samplingtrue
temperature0.70
top_k20
top_p0.90
min_p0.05
repeat_penalty1.00
context_size4096
show_everything_in_consolefalse
CategoryPrompt Manager

Inputs (15)

NameTypeDefaultDescription
modeloptCOMBOQwen3.5-9B-UD-Q4_K_XL.ggufSelect model to use (local models listed first, then HuggingFace models) Download sizes: UD-Q4_K_XL ~6GB | Q8_0 ~9.5GB | UD-Q8_K_XL ~13GB
image2optIMAGEConnect an image (required for 'Analyze Image' and 'Analyze Image with Prompt' modes)
image3optIMAGEConnect an image (required for 'Analyze Image' and 'Analyze Image with Prompt' modes)
image4optIMAGEConnect an image (required for 'Analyze Image' and 'Analyze Image with Prompt' modes)
image5optIMAGEConnect an image (required for 'Analyze Image' and 'Analyze Image with Prompt' modes)
system_prompt_modeoptCOMBOappendreplace: the text below fully replaces the default LLM instructions append: the text below is added after the default LLM instructions
system_promptoptSTRINGCustom LLM Instructions (leave empty to use default) The default instructions are designed for generating detailed and imaginative prompts for text-to-image generation.
use_model_default_samplingoptBOOLEANtrueUse the model's default sampling parameters (overrides temperature, top_p, etc)
temperatureoptFLOAT0.700–2Controls randomness (0.0 = deterministic, 2.0 = very random)
top_koptINT200–100Sample from top K most likely tokens (0 = disabled)
top_poptFLOAT0.900–1Nucleus sampling: consider tokens with top_p probability mass
min_poptFLOAT0.050–1Minimum probability threshold relative to top token
repeat_penaltyoptFLOAT1.001–2Penalty for repeating tokens (1.0 = no penalty)
context_sizeoptINT4096512–32768Context size (increase for vision models or large prompts)
show_everything_in_consoleoptBOOLEANfalsePrint system prompt, user prompt, thinking process, and raw model response to console

Outputs (1)

NameTypeDescription
optionsOPTIONS