Nodes/ComfyUI_LC123_nodes/LC Detail Daemon (BETA)
ComfyUI Node

LC Detail Daemon (BETA)

Adds fine detail by telling the model there is a little less noise than there really is, in a window of the steps. Same scale and window as the original Detail Daemon, plus keep structure mode, and no CFG maths.

By lonecatone23·Created 2 months ago·Updated about 5 hours ago· 26
LC Detail Daemon (BETA)
  • sampler
  • sigmas
  • SAMPLER
◄amount1.00►
◄start0.15►
◄end0.85►
◄peak0.00►
◄exponent0.00►
◄modelook only►
CategoryLC123/sampling

Inputs (8)

NameTypeDefaultDescription
samplerSAMPLERFrom KSamplerSelect, LC Sigmas or ClownSampler (any SAMPLER).
amountFLOAT1.00-5–5How much detail. The same scale as the original Detail Daemon: 0.25 = the model is told 2.5% less noise at the peak. Negative = smoother. Not multiplied by CFG. Krea 2 / Krealism: look only 1.0, start 0.15, end 0.85, peak 0, exponent 0 (the defaults). classic and keep structure hit harder: start lower. At 16+ steps use less. Samplers that add their own noise (dpmpp_2m_sde, euler_ancestral, er_sde, res_* / ClownSampler with eta): lower it, or it can leave specks.
startFLOAT0.150–1The step where it begins: 0 = first step, 1 = last. Early steps shape folds and fabric.
endFLOAT0.850–1The step where it ends. Leave the last steps alone, or noise can stay in the image.
peakFLOAT0.000–1Where in the window it is strongest (the original's bias): 0 = right at start, on the noisy steps, where it shapes things (works best); higher = later, a gentler sharpen.
exponentFLOAT0.000–10The ramp's curve: 0 = flat, full strength across the whole window (works best); 1 = a smooth bump; above 1 = narrower.
modeCOMBOlook onlylook only = the model sees less noise, but its answer is rescaled to the real noise: crisp, without big reshuffles. classic = the original Detail Daemon (more detail, small things shift). keep structure = classic, but only the fine part of the change is kept, so the layout stays (two model calls per step in the window).
sigmasoptSIGMASOptional, for the graph only: shows where your steps land.

Outputs (1)

NameTypeDescription
SAMPLERSAMPLER—