Dazzle Bong Sampler (DazzleNodes)

Dazzle Bong Sampler is the pack's second full sampler - model, conditioning, and latent in, LATENT out, just like the flagship DazzleKSampler. The difference is the menu. Where the main node offers 119 solver choices, Bong trims it to the eight res family methods (res_2m, res_3m, res_2s, res_3s, and their _sde variants), defaulting to res_2s_sde. It's a port of RES4LYF's BongSampler, and it reads like the author's "here's the shortlist I actually use" node - no 107-method dropdown to get lost in.

Why would you pick it over the flagship? Mostly taste and signal-to-noise. The res family is the heart of RES4LYF's solver lineup - the explicit RK methods people actually name in workflow shares (res_2m, res_2s) - and a restricted list is a feature when you're iterating and don't want to accidentally land on some exotic method you don't recognize. Everything else is standard: steps, cfg (default 5.5), denoise, seed, scheduler (default beta57), and the optional positive / negative / latent_image sockets.

The input that's actually new

latent_role works the same way it does on the flagship: auto inspects the incoming latent dict and decides whether the upstream tensor is being used as init, as pure noise, or as layered noise-plus-init. That's the shared shaped-noise passthrough feature running through the whole Dazzle family, and here it's fully live - Bong consumes the latent directly, so the role choice actually does something (unlike the SAMPLER-emitting nodes where the widget is advisory). Leave it on auto unless you're deliberately forcing a path.

Output is a single output (LATENT) - no denoised, no options, just the result. Feed it to a VAE Decode and you're done.

Installing it

Same pack as every Dazzle sampler:

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

Restart ComfyUI; it appears under RES4LYF/samplers. ComfyUI Manager users search "Dazzle KSampler". No pip dependencies, no model downloads, AGPL-3.0 inherited from RES4LYF.

Where people get burned

The honest caveat is the opposite of the other nodes: here the trap is what's not available. Want dpmpp_2m or one of the implicit solvers? Not on this list - this node is the res family or nothing, so a workflow copied from someone using the flagship won't load onto Bong unchanged. The res_2s_sde default is ancestral (SDE), which means it doesn't settle the way res_2m does - same seed, same result isn't guaranteed. If you're A/B testing settings, switch to a converging res_2m/res_3m variant so you're actually comparing one thing at a time.