Nodes/RES4LYF/ClownOptions Detail Boost
ComfyUI Node Runs on cloud

ClownOptions Detail Boost

Add detail during sampling, on a step window

By ClownsharkBatwing·Created 2 years ago·Updated 22 days ago· 1,222
ClownOptions Detail Boost
  • weights
  • etas
  • options
  • options
weight1.00
methodmodel
modehard
eta0.50
start_step3
end_step10

Detail Boost is RES4LYF's answer to "the image is fine but a bit soft." It's an options module that pushes extra fine detail into a generation during sampling, rather than as a post-process sharpen. You wire its OPTIONS output into a RES4LYF sampler, aim it at a window of steps, and it works detail in over that range. The author introduced it alongside the pack's faceswap and style-transfer features, where flat, plastic output is the enemy and a little extra structure goes a long way.

The reason to do this at sampling time instead of after: sharpening a finished image just amplifies what's already there, edges and artifacts included. Adding detail while the model is still resolving lets the model decide where the new structure goes, so you get plausible texture instead of a crunchy overlay.

How it works

Over the step range you set, the node applies a detail-boosting operation controlled by a weight and a mode. The method picks where the boost is injected - into the sampler's behavior or into the model call - and mode picks the noise/scaling curve used. Confining it to a step window matters: boosting during the mid-to-late steps adds fine detail without disturbing the composition that got locked in early.

The inputs and outputs that matter

  • weight (FLOAT, default 1) - how much boost. This is the main dial; higher is more aggressive. Overdo it and you get noise and crunch, so nudge it.
  • method (default model) - where the boost is applied: model, model_alpha, or the sampler* variants. model is the sensible starting point.
  • mode (default hard) - the scaling curve: hard, lorentzian, soft, softer, exp, and others, running from strong to gentle.
  • start_step / end_step (INT, default 3 and 10) - the window the boost acts over. The defaults deliberately skip the first couple of steps (leave composition alone) and stop before the very end. Widen or shift this to move where the detail lands.
  • eta (FLOAT, default 0.5) - the noise amount the boost uses.

The single output is options (OPTIONS) - feed it to your sampler, or chain it with other ClownOptions nodes via their options input.

Installing it

ComfyUI Manager: search RES4LYF, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ClownsharkBatwing/RES4LYF/
cd RES4LYF
pip install -r requirements.txt

pip install in the venv or portable Python, restart, hard-refresh (F5). No downloads.

Where people get tripped up

Weight is easy to overcook. A little detail boost reads as crispness and realism; too much reads as noise, halos, or a busy, over-textured surface. Start at or below the default weight, look at the result at 100%, and creep up only if it genuinely needs it - this is a seasoning node, not a slider to max out.

The step window is the other half of the skill. Start the boost too early and it fights the composition; run it all the way to the final step and it can leave a gritty finish. The 3–10 default is a reasonable mid-late window for a ~20–30 step run; if you change your step count, move the window to match so the boost still lands in the detail phase rather than the layout phase. And remember it's an OPTIONS module - it does nothing on its own, it needs a RES4LYF sampler to read it.

CategoryRES4LYF/sampler_options

Inputs (9)

NameTypeDefaultDescription
weightFLOAT1.00-100–100Set to positive values to create a sharper, grittier, more detailed image. Set to negative values to soften and deepen the colors.
methodCOMBOmodelDetermines whether the sampler or the model underestimates the noise level.
modeCOMBOhardChanges the steps where the effect is greatest. Most affect early steps, sinusoidal affects middle steps.
etaFLOAT0.50-100–100The strength of the effect of the noise_scaling_mode. Linear ignores this parameter.
start_stepINT30–10000
end_stepINT10-1–10000
weightsoptSIGMAS
etasoptSIGMAS
optionsoptOPTIONS

Outputs (1)

NameTypeDescription
optionsOPTIONS