Nodes/ComfyUI-Prompt-Rewriter/Prompt Rewriter Options
ComfyUI Node

Prompt Rewriter Options

A ComfyUI node in Prompt Manager with 18 inputs and 1 output.

By BigStationW·Created 9 months ago·Updated 5 months ago· 20
Prompt Rewriter Options
  • image_1
  • image_2
  • image_3
  • image_4
  • image_5
  • options
modelNo models available
gpu_layers
enable_thinkingtrue
context_size4096
max_tokens4096
flash_attentiontrue
system_prompt
use_model_default_samplingfalse
temperature0.80
top_p0.95
top_k40
min_p0.05
repeat_penalty1.00
CategoryPrompt Manager

Inputs (18)

NameTypeDefaultDescription
modelCOMBONo models availableSelect model to use. Models with ⬇ will be downloaded automatically.
gpu_layersSTRINGGPU layer distribution. Examples: • empty -> All layers go to the first GPU (default) • gpu0:0.7 -> 70% to GPU:0, 30% to CPU • gpu0:0.5, gpu1:0.4 -> 50% GPU:0, 40% GPU:1, 10% CPU
enable_thinkingBOOLEANtrueEnable thinking/reasoning mode (model thinks before answering)
context_sizeINT4096256–1000000Context window size (total tokens for input + output). Higher values use more VRAM.
max_tokensINT40961–1000000Maximum tokens to generate
flash_attentionBOOLEANtrueEnable Flash Attention for faster inference (recommended: enabled)
system_promptSTRINGOverride the default system prompt
use_model_default_samplingBOOLEANfalseUse the model's default sampling parameters (overrides temperature, top_p, etc)
temperatureFLOAT0.800–2Temperature for generation (higher = more creative, lower = more focused)
top_pFLOAT0.950–1Top-p (nucleus) sampling
top_kINT400–200Top-k sampling (0 = disabled)
min_pFLOAT0.050–1Min-p sampling threshold
repeat_penaltyFLOAT1.000–2Repetition penalty (1.0 = no penalty, higher = less repetition)
image_1optIMAGEOptional image input for VLM models (image 1 of 5). Requires mmproj file.
image_2optIMAGEOptional image input for VLM models (image 2 of 5). Requires mmproj file.
image_3optIMAGEOptional image input for VLM models (image 3 of 5). Requires mmproj file.
image_4optIMAGEOptional image input for VLM models (image 4 of 5). Requires mmproj file.
image_5optIMAGEOptional image input for VLM models (image 5 of 5). Requires mmproj file.

Outputs (1)

NameTypeDescription
optionsOPTIONS