Nodes/ComfyUI-DrawThings-gRPC/Draw Things Sampler
ComfyUI Node

Draw Things Sampler

A ComfyUI node in DrawThings with 63 inputs and 1 output.

By Jokimbe·Created about a year ago·Updated 4 months ago· 31
Draw Things Sampler
  • model
  • image
  • mask
  • lora
  • control_net
  • upscaler
  • refiner
  • hints
  • IMAGE
settingsBasic
serverlocalhost
port7859
use_tlstrue
strength1.00
seed0
seed_modeScaleAlike
width512
height512
steps20
num_frames14
cfg4.5
cfg_zero_starfalse
cfg_zero_star_init_steps0
speed_uptrue
guidance_embed4.5
sampler_nameDPM++ 2M AYS
stochastic_sampling_gamma0.30
res_dpt_shifttrue
shift1.00
batch_size1
fps5
motion_scale127
guiding_frame_noise0.02
start_frame_guidance1.0
causal_inference0
causal_inference_pad0
clip_skip1
sharpness0.6
mask_blur1.5
mask_blur_outset0
preserve_originaltrue
high_res_fixfalse
high_res_fix_start_width448
high_res_fix_start_height448
high_res_fix_strength0.70
tiled_decodingfalse
decoding_tile_width640
decoding_tile_height640
decoding_tile_overlap128
tiled_diffusionfalse
diffusion_tile_width512
diffusion_tile_height512
diffusion_tile_overlap64
tea_cachefalse
tea_cache_start5
tea_cache_end2
tea_cache_threshold0.20
tea_cache_max_skip_steps3
separate_clip_lfalse
clip_l_text
separate_open_clip_gfalse
open_clip_g_text
positive
negative
CategoryDrawThings

Inputs (63)

NameTypeDefaultDescription
settingsCOMBOBasic3 options: Basic, Advanced, All
serverSTRINGlocalhostThe IP address of the Draw Things gRPC Server.
portSTRING7859The port that the Draw Things gRPC Server is listening on.
use_tlsBOOLEANtrue
modelDT_MODELThe model used for denoising the input latent.
strengthFLOAT1.000–1When generating from an image, a high value allows more artistic freedom from the original. 1.0 means no influence from the existing image (a.k.a. text to image).
seedINT0-1–4294967295The random seed used for creating the noise.
seed_modeCOMBOScaleAlike4 options: Legacy, TorchCpuCompatible, ScaleAlike, NvidiaGpuCompatible
widthINT512128–8192
heightINT512128–8192
stepsINT201–150The number of steps used in the denoising process.
num_framesINT141–201
cfgFLOAT4.50–50The Classifier-Free Guidance scale balances creativity and adherence to the prompt. Higher values result in images more closely matching the prompt however too high values will negatively impact quality.
cfg_zero_starBOOLEANfalse
cfg_zero_star_init_stepsINT00–50
speed_upBOOLEANtrue
guidance_embedFLOAT4.50–50
sampler_nameCOMBODPM++ 2M AYSThe algorithm used when sampling, this can affect the quality, speed, and style of the generated output.
stochastic_sampling_gammaFLOAT0.300–1
res_dpt_shiftBOOLEANtrue
shiftFLOAT1.000.1–16
batch_sizeINT11–4
fpsINT51–30
motion_scaleINT1270–255
guiding_frame_noiseFLOAT0.020–1
start_frame_guidanceFLOAT1.00–25
causal_inferenceINT00–129Set to 0 to disable causal inference
causal_inference_padINT00–129
clip_skipINT11–23
sharpnessFLOAT0.60–30
mask_blurFLOAT1.50–50
mask_blur_outsetINT0-100–1000
preserve_originalBOOLEANtrue
high_res_fixBOOLEANfalse
high_res_fix_start_widthINT448128–2048
high_res_fix_start_heightINT448128–2048
high_res_fix_strengthFLOAT0.700–1
tiled_decodingBOOLEANfalse
decoding_tile_widthINT640128–2048
decoding_tile_heightINT640128–2048
decoding_tile_overlapINT12864–1024
tiled_diffusionBOOLEANfalse
diffusion_tile_widthINT512128–2048
diffusion_tile_heightINT512128–2048
diffusion_tile_overlapINT6464–1024
tea_cacheBOOLEANfalse
tea_cache_startINT50–1000
tea_cache_endINT2-151–150A positive value corresponds to last step tea_cache should be appplied. Use a negative value to specify steps from the end+1. IE: -1 is the last step, -2 is the second to last step, and so on.
tea_cache_thresholdFLOAT0.200–1
tea_cache_max_skip_stepsINT31–50
separate_clip_lBOOLEANfalse
clip_l_textSTRING
separate_open_clip_gBOOLEANfalse
open_clip_g_textSTRING
imageoptIMAGE
maskoptMASKA black/white image where black areas will be kept and the rest will be regenerated according to your strength setting.
positiveoptSTRINGThe conditioning describing the attributes you want to include in the image.
negativeoptSTRINGThe conditioning describing the attributes you want to exclude from the image.
loraoptDT_LORA
control_netoptDT_CNET
upscaleroptDT_UPSCALER
refineroptDT_REFINER
hintsoptDT_HINTSHint images, or control inputs, are used with ControlNets and certain other models, like Kontext, to help guide the generation.

Outputs (1)

NameTypeDescription
IMAGEIMAGE