Nodes/RES4LYF/ClownGuidesAB
ComfyUI Node Runs on cloud

ClownGuidesAB

Drive a generation with two latent guides at once

By ClownsharkBatwing·Created 2 years ago·Updated 18 days ago· 1,222
ClownGuidesAB
  • guide_A
  • guide_B
  • mask_A
  • mask_B
  • weights_A
  • weights_B
  • guides
guide_modeepsilon
channelwise_modefalse
projection_modefalse
weight_A0.75
weight_B0.75
cutoff_A1.00
cutoff_B1.00
weight_scheduler_Abeta57
weight_scheduler_Bconstant
start_step_A0
start_step_B0
end_step_A15
end_step_B15
invert_masksfalse

ClownGuidesAB is the workhorse of RES4LYF's latent-guidance system: two reference latents, A and B, each with its own strength, schedule, step window, and mask, feeding one generation. This is the node behind the pack's dual-guide workflows - the "composition guide plus style guide" setups, and the masked dual-guide unsampling examples in the docs, where one reference shapes the layout and another shapes the look.

Every RES4LYF guide builds a GUIDES bundle that plugs into the sampler's guides input and steers denoising toward your references. Most guides carry one reference. ClownGuidesAB carries two, independently controlled, which is what lets you say "follow this image's composition here, and that image's content there." If you're doing image-to-image, unsampling/resampling, or region-masked guidance, this is usually the node in the middle of it.

How it works

You give it up to two guide latents and choose a guide_mode that decides how the references act on the sampling process. The modes range from straightforward (epsilon, the default, and data) to the more exotic (flow, sync, lure, inversion, pseudoimplicit, and self-refine variants), each a different mathematical relationship between the reference and the denoising trajectory. A and B get separate weights, schedulers, and start/end steps, so you can have guide A dominate early for composition and guide B come in later for detail, each confined to its own mask.

The inputs and outputs that matter

  • guide_A / guide_B (LATENT, optional) - your two references. Use one for single-guide work, both for dual.
  • guide_mode (enum, default epsilon) - how the guides act. Start with epsilon; it's the sane default. The others are for specific tricks and unsampling flows.
  • weight_A / weight_B (FLOAT, default 0.75) - strength of each guide. Your primary dials.
  • weight_scheduler_A / _B (enum; A defaults to beta57) - per-guide scheduling across steps.
  • start_step_A/B / end_step_A/B (INT) - each guide's active window; this is how you stage composition-then-detail.
  • mask_A / mask_B (MASK, optional) plus invert_masks - confine each guide to a region.

Output is a guides (GUIDES) bundle for the sampler (or another guide node).

How to install it

ComfyUI Manager: search RES4LYF, install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/ClownsharkBatwing/RES4LYF/, pip install -r requirements.txt in the venv (portable ComfyUI: embedded python's pip), restart, hard-refresh F5. The pack's example dual-guide and unsampling workflows are the fastest way to see this wired correctly. Nested sampler menu: rgthree-comfy with "Auto Nest Subdirectories" on.

Common issues

The guide_mode enum is where people get lost - ten options with almost no in-app explanation, which is the RES4LYF documentation gap in one dropdown. Don't sweep all ten hoping one clicks. Stay on epsilon unless a workflow you're copying tells you otherwise; the exotic modes (inversion, pseudoimplicit, sync/lure) are tied to specific unsampling and refinement techniques and will look broken if you use them outside those contexts.

Two more. Guides interact with the sampler's noise settings, so for unsampling-style work you generally set the sampler to unsample/resample mode and often CFG 1 - a guide "doing nothing" is frequently a sampler-mode mismatch, not a bad weight. And the scheduler caveat is live here too: bong_tangent runs its own sigma shift and ignores the workflow's, so if a per-guide scheduler behaves strangely, fall back to beta57 or constant. Fixed seed while you tune, always - with two guides and this many knobs, seed variance will otherwise drown out what you changed.

CategoryRES4LYF/sampler_extensions

Inputs (20)

NameTypeDefaultDescription
guide_modeCOMBOepsilonCompletely undocumented guide mode.
channelwise_modeBOOLEANfalse
projection_modeBOOLEANfalse
weight_AFLOAT0.75-100–100Set the strength of the guide.
weight_BFLOAT0.75-100–100Set the strength of the guide_bkg.
cutoff_AFLOAT1.000–1Disables the guide for the next step when the denoised image is similar to the guide. Higher values will strengthen the effect.
cutoff_BFLOAT1.00-100–100Disables the guide for the next step when the denoised image is similar to the guide. Higher values will strengthen the effect.
weight_scheduler_ACOMBObeta5712 options: constant, simple, sgm_uniform, karras, exponential, ddim_uniform, +6
weight_scheduler_BCOMBOconstant12 options: constant, simple, sgm_uniform, karras, exponential, ddim_uniform, +6
start_step_AINT00–10000
start_step_BINT00–10000
end_step_AINT15-1–10000
end_step_BINT15-1–10000
invert_masksBOOLEANfalse
guide_AoptLATENT
guide_BoptLATENT
mask_AoptMASK
mask_BoptMASK
weights_AoptSIGMAS
weights_BoptSIGMAS

Outputs (1)

NameTypeDescription
guidesGUIDES