Legacy_ClownsharKSampler

If ComfyUI is showing you a node called Legacy_ClownsharKSampler, you're almost certainly loading an old workflow, and it will still run - that's the whole reason it exists. RES4LYF's policy is to keep previous generations available with "Legacy" prepended rather than break your saved graphs. There are actually two tiers now (Legacy and Legacy2), and this plain-Legacy node is the oldest of the ClownsharKSampler lineage. The current one is ClownsharKSampler_Beta. Nothing wrong with running this; just don't expect it to match newer tutorials exactly.

What it is

It's the same core idea as every ClownsharKSampler: an all-in-one node that fuses the solver picker with the thing that runs it, so model, conditioning, scheduler, steps and CFG all live on one node. RES4LYF is the pack (by ClownsharkBatwing) that became the default home for sampler tuning on flow-matching models - Flux, SD3.5, Wan, Chroma - because its high-order res_ solvers suit the near-straight denoising trajectory those models use, where the old Karras defaults fall apart.

The one thing that visibly separates this older node from the newer ones is that shift lives right on it. Later generations moved shift out to the ModelSampling nodes; here it's a first-class widget, which is honestly convenient if you don't want an extra node in the graph.

The inputs that matter

You need a model and a latent_image. After that:

• sampler_name (default res_2m) - the solver. This legacy node offers a shorter list (29 choices) than the current ones, but res_2m is here and it's still the right everyday pick.
• scheduler (default beta57) - the sigma schedule; beta57 is the pack's signature and a solid default for flow-matching models.
• shift (default 3) and base_shift (default 0.85) - the timestep shift. shift matches the ModelSampling "shift" (or Flux "max_shift"); base_shift is Flux-only. Set either to -1 to disable it from inside the node. On a straight trajectory, shift is roughly where scheduler tuning went, so it's worth playing with.
• cfg (default 5), steps (default 30), denoise/denoise_alt - behave like KSampler. Drop CFG to 1.0 on guidance-distilled Flux.
• truncate_conditioning (default true) - a conditioning-handling toggle from this era of the node. If you're rebuilding an old screenshot workflow, match whatever it showed.

eta controls SDE noise injection; implicit_steps/implicit_sampler_name add implicit refinement (leave at 0 unless you want it). Optional sigmas overrides the scheduler; guides takes RES4LYF guide images. Outputs are output (finished latent → VAE Decode) and denoised.

Installing it

ComfyUI Manager: search custom nodes for RES4LYF, install, restart. Manually, from your ComfyUI folder:

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

Run pip install inside the venv (or portable Python), restart, hard-refresh F5. No models to download.

Common issues

Because this is the original node, its layout doesn't line up with the current readme, and that mismatch is the main source of confusion. If you're recreating a previous-version workflow from screenshots, the readme's own advice applies: you may need to delete and recreate the sampler and guide nodes and re-enter the settings by hand, and if the old screenshot used truncate_conditioning=true you'd add that in the newer node's extra_options box. If nothing loads, the pack isn't installed. And if you're starting from scratch rather than restoring something, save yourself the friction and use ClownsharKSampler_Beta.