Dazzle Clown Sampler (DazzleNodes)

If the pack's Advanced node is a sampler factory with every dial exposed, Dazzle Clown Sampler is the same idea stripped to the essentials. It's a port of RES4LYF's ClownSampler, and it does exactly one thing: emit a SAMPLER object you can hand to a downstream runner like SamplerCustomAdvanced or this pack's DazzleSharkSampler. No model socket, no conditioning, no latent - the actual sampling happens elsewhere.

You'd reach for this over the Advanced node for a simple reason: it's clean. The full RES4LYF tuning surface can be overwhelming, and Clown keeps it to the four controls that matter most. If you want a sensible RES4LYF solver wired into your graph without wading through twenty advanced dials, this is the low-friction way in.

The inputs that matter

• sampler_name - the full 119-method RES4LYF list, default res_2m. This is the whole point: res_2m, res_3m, dpmpp_2m, and the rest, ready for flow-matching models.
• eta - per-step ancestral/SDE noise (default 0.5). Drop it toward 0 for more deterministic, more convergent behavior.
• seed - feeds the per-step noise machinery inside the eventual run.
• bongmath - the RES4LYF math toggle, on by default. It's the pack's signature one-line upgrade; leave it on unless you know why you're turning it off.

There's also a latent_role widget, but read the tooltip before you trust it: it's advisory. This node returns a SAMPLER and doesn't consume a latent directly, so the widget is shown for UI consistency with the rest of the Dazzle samplers. The actual noise-vs-init dispatch happens in whichever node runs the sampler. Optional guides and options sockets exist for RES4LYF-ecosystem wiring; the single output is sampler (SAMPLER).

Installing it

It ships in the same pack as all the Dazzle samplers:

cd ComfyUI/custom_nodes
git clone https://github.com/DazzleNodes/ComfyUI-DazzleKSampler.git

Restart ComfyUI and it shows up under RES4LYF/samplers. Or install via ComfyUI Manager by searching "Dazzle KSampler". No pip dependencies or model files.

Where people get burned

The recurring confusion is expecting a picture out of it - it returns a SAMPLER, not a latent, so it can't connect to a VAE Decode. And remember that latent_role here is cosmetic: don't set it expecting it to steer this node's behavior, because this node doesn't sample anything. Pair it with a runner node and it's genuinely pleasant to use; on its own it's just a config object waiting for a job.