Legacy_ClownsharKSamplerGuides
Bundling img2img guidance and schedule for the old sampler
- latent_guide
- latent_guide_inv
- latent_guide_mask
- latent_guide_weights
- GUIDES
Part of the pre-rewrite generation of RES4LYF nodes - same family as Legacy_ClownSampler - and, like that one, kept around specifically so old saved workflows keep loading rather than breaking after the pack's update. Legacy_ClownsharKSamplerGuides packages up img2img-style guidance settings plus the scheduler/steps/denoise trio into a single GUIDES object, which the legacy sampler chain consumes.
If you're not sure whether you need it: on a fresh workflow using the current ClownsharKSampler, you don't - that node handles guidance and scheduling internally. This node is what you'll find wired into older graphs instead.
How it works
latent_guide (and its inverse counterpart, latent_guide_inv) let you steer the generation toward a reference latent - classic img2img guidance, but exposed as a first-class RES4LYF concept rather than bolted on. guide_mode picks how that reference gets blended in: hard_light, mean_std, mean, std, or blend, each a different way of merging your guide latent with what the sampler is generating. latent_guide_weight sets how strongly it pulls, and latent_guide_mask restricts the effect to part of the image. Separately, scheduler, steps, and denoise set the sigma schedule this guide config runs against - including beta57, RES4LYF's own signature schedule, and bong_tangent right alongside the stock ComfyUI options.
Worth knowing if you pick bong_tangent: it applies its own internal sigma shift and ignores whatever shift value the rest of your workflow is set to. If you've dialed in a shift on a ModelSamplingAuraFlow-style node and it appears to do nothing, check whether your scheduler here is bong_tangent before assuming the shift setting is broken.
The inputs and outputs that matter
guide_mode(defaultblend) - how the reference latent gets merged in.blendis the safest starting point;hard_lightis the most aggressive.latent_guide_weight- how strongly the guide pulls the result toward itself.scheduler- the sigma schedule, includingbeta57andbong_tangent(the latter with the shift caveat above).steps/denoise- the usual step count and denoise strength, scoped to this guide config.- Optional
latent_guide,latent_guide_inv,latent_guide_mask,latent_guide_weights- the actual reference latent(s) and an optional mask to restrict where guidance applies. - Output:
GUIDES- feed this into the matching legacy sampler in your chain.
How to install it
Via ComfyUI Manager: search "RES4LYF" and install. Manually, from ComfyUI/custom_nodes:
git clone https://github.com/ClownsharkBatwing/RES4LYF/
cd RES4LYF
pip install -r requirements.txt
Activate your venv first if you use one; on portable installs, swap pip for python_embedded\Scripts\pip.exe. Restart ComfyUI afterward.
Common issues & troubleshooting
shift seems to do nothing. As above - this is a known trap with the bong_tangent scheduler specifically, which overrides your workflow's shift value with its own. Switch schedulers to confirm before concluding shift is broken elsewhere in your graph.
Guidance is too strong or barely visible. Start by adjusting latent_guide_weight before touching guide_mode - it's the more predictable of the two knobs. Once you have a weight that feels close, experiment with guide_mode to change the character of the blend rather than just its strength.
You're not sure if you should still be using this node. If you're not loading an older saved workflow, you probably shouldn't be - the current ClownsharKSampler folds this functionality in directly and is the maintained path going forward. This node exists for backward compatibility, not as the recommended starting point.
Inputs (10)
| Name | Type | Default | Description |
|---|---|---|---|
| guide_mode | COMBO | blend | The mode used. |
| latent_guide_weight | FLOAT | 0.00-100–100 | — |
| scheduler | COMBO | beta57 | 12 options: constant, simple, sgm_uniform, karras, exponential, ddim_uniform, +6 |
| steps | INT | 301–10000 | — |
| denoise | FLOAT | 1.00-100–100 | — |
| rescale_floor | BOOLEAN | false | If true, latent_guide_weight(s) primarily affect the masked areas. If false, they control the unmasked areas. |
| latent_guideopt | LATENT | — | |
| latent_guide_invopt | LATENT | — | |
| latent_guide_maskopt | MASK | — | |
| latent_guide_weightsopt | SIGMAS | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| GUIDES | GUIDES | — |