Legacy2_ClownsharKSamplerGuide
Steer a generation with a guide latent
- guide
- guide_mask
- guide_weights
- guides
This node is how you feed a RES4LYF sampler a guide - a latent that steers the generation toward a particular composition, color layout, or style without hard-locking it the way a fixed init image would. It's one of the pack's most-loved features. A regular community description puts it well: it sits somewhere in the neighborhood of img2img at high denoise, or unsampling / RF-inversion / an ad-hoc controlnet for models that don't have one. Great for guiding composition or color, or just nudging outputs away from the "person standing dead-center posing for the camera" default.
Heads up on the name: ComfyUI shows this as Legacy2_ClownsharKSamplerGuide, so it belongs to an older generation kept alive for existing workflows. The current guide nodes have different names, but the concept is identical and this still works.
How it works
You give it a guide latent and a guide_mode, and it produces a GUIDES object that plugs into a Clown sampler's guides input. The sampler then blends your guide into the denoising trajectory according to the mode and weight you chose, tapering off at a step you set so the model has room to finish on its own. This is a single-guide node - one guide, optionally masked. (If you want a foreground/background pair, that's the plural ClownsharKSamplerGuides.)
The inputs that matter
- guide_mode (default
epsilon_projection) - how the guide is injected. There are 17 modes (epsilon,epsilon_projection,blend,hard_light,mean_std, plus dynamic-mean and background variants).epsilon-family modes are the common starting point;blendand the projection modes give softer or more literal adherence. This is the knob with the biggest effect on feel, so it's the one to experiment with. - guide_weight (default 0.75) - how strongly the guide pulls. Higher tracks the guide more closely; lower lets the prompt lead.
- guide_end_step (default 15) - the step to stop guiding. Guide hard early for structure, release for detail. Ending too late can lock in the guide's texture; too early barely registers.
- guide_weight_scheduler (default
beta57) - schedules how the weight changes over the guided steps rather than holding it flat.
Optional inputs: guide (the LATENT you're steering with - encode an image through a VAE to make one), guide_mask (a MASK to confine the guide to a region), and guide_weights (a SIGMAS curve for full manual control of the weight schedule). The single output is guides (GUIDES).
Installing it
ComfyUI Manager: search RES4LYF, install, restart. Manually:
cd custom_nodes
git clone https://github.com/ClownsharkBatwing/RES4LYF/
cd RES4LYF
pip install -r requirements.txt
pip install inside the venv (or portable Python), restart, hard-refresh F5. No models to download - but you'll need a VAE Encode in your graph to turn a reference image into the guide latent.
Common issues
The most common "it's not working" is simply that the GUIDES output isn't wired into the sampler's guides input, or that guide_weight is too low to see. Turn it up, or drop the guide_end_step later, to confirm the guide is actually biting - then dial back. If you're doing style transfer, the readme's advice is worth heeding: try with the guide off first to confirm your prompt isn't doing the heavy lifting, because a strong style word in the prompt can mask what the guide is contributing. And since this is a Legacy2 node, expect its exact settings to differ slightly from the newest tutorials. If it won't load, RES4LYF isn't installed.
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| guide_mode | COMBO | epsilon_projection | Recommended: epsilon or mean/mean_std with sampler_mode = standard, and unsample/resample with sampler_mode = unsample/resample. Epsilon_dynamic_mean, etc. are only used with two latent inputs and a mask. Blend/hard_light/mean/mean_std etc. require low strengths, start with 0.01-0.02. |
| guide_weight | FLOAT | 0.75-100–100 | Set the strength of the guide. |
| guide_weight_scale | FLOAT | 1.000–1 | Disables the guide for the next step when the denoised image is similar to the guide. Higher values will strengthen the effect. |
| guide_weight_scheduler | COMBO | beta57 | 12 options: constant, simple, sgm_uniform, karras, exponential, ddim_uniform, +6 |
| guide_end_step | INT | 151–10000 | — |
| guideopt | LATENT | — | |
| guide_maskopt | MASK | — | |
| guide_weightsopt | SIGMAS | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| guides | GUIDES | — |