SimpleSampler

ComfyUI's stock KSampler hands you every dial separately - sampler name, scheduler, steps, CFG, denoise - and expects you to know which combination goes with which kind of model. SimpleSampler takes the opposite approach: instead of exposing those dials individually, it bakes four whole recipes into a single sampler dropdown, so picking "SDXL Turbo" sets you up with the sampler and scheduler that model family actually needs, rather than making you remember it.

That trade-off is the entire point of this node, and it's worth being honest about both sides of it. You lose fine control - there's no steps or CFG input here at all, so if you want to hand-tune those, this isn't the node for that. What you gain is protection from a genuinely common mistake: distilled and LoRA-accelerated models (LCM, SDXL Turbo, and their relatives like Lightning or Hyper-SD) need completely different settings from a normal checkpoint, and running them at normal-model settings - or vice versa - produces garbage, not just slightly-off results. The community shorthand for this is blunt: every time a model description says "Lightning," "Turbo," or "LCM," the settings are completely different from the standard version, full stop.

The four presets, and what's actually behind them

• Normal - euler / Normal - uni_pc - for ordinary, non-distilled checkpoints. Use whichever sampler you'd normally reach for; these are just the two the node offers as "standard" options.
• LCM Lora - lcm - for a checkpoint running an LCM LoRA. LCM needs its own dedicated sampler (which is what this preset selects) and is only correct at very low CFG - typically 1 to 2 - over just 4 to 8 steps. Run it at normal CFG and step counts and you'll get noise or badly degraded output; that's not this node misbehaving, that's what happens when an LCM-tuned model runs outside the narrow range it was distilled for.
• SDXL Turbo - dpmpp_sde karras - for actual SDXL Turbo checkpoints specifically, not just "any SDXL model." Turbo wants roughly CFG 2 and only 1 to 4 steps - again, a completely different regime from a normal SDXL run, and this preset is choosing the sampler/scheduler pairing that regime expects.

The inputs and outputs that matter

• model - your loaded checkpoint (or checkpoint + LCM LoRA already applied).
• sampler - the four-way preset above; pick the one matching what your model actually is.
• positive / negative - CONDITIONING, same as any KSampler.
• latents - your starting LATENT, from an empty latent for txt2img or an encoded image for img2img.
• mode - txt2img or img2img; this is what tells the node whether to treat your incoming latent as pure noise to build from scratch or as a starting image to partially redenoise.
• seed (optional) - fix it for reproducible results.

One output: LATENT, ready for a VAE decode or straight into Chibi's own SaveImages (which can decode a latent for you directly).

How to install it

• ComfyUI Manager: search "ComfyUI-Chibi-Nodes", install, restart.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/chibiace/ComfyUI-Chibi-Nodes/, restart.

No models to download for the node itself - though obviously the model you load into it (a base checkpoint, or one with an LCM LoRA applied for that preset) is on you to source separately. No extra dependencies. It appears under Chibi-Nodes once installed.

Common issues

The single biggest mistake with this node is a preset/model mismatch: picking "SDXL Turbo" while running a normal SDXL checkpoint, or "Normal - euler" while running an actual distilled model. Because none of the usual dials (steps, CFG) are visible here to sanity-check against, that mismatch is easy to make and confusing to diagnose - the output just looks bad, with nothing in the node's UI hinting at why. If your results look like noise or heavily degraded structure, the first thing to check isn't your prompt, it's whether the sampler preset actually matches the model you loaded.

The other limitation worth internalizing rather than fighting: if you need to hand-tune steps, CFG, or denoise strength - which you will, eventually, for anything beyond a quick preset-driven run - this node genuinely can't do it. Drop back to a stock KSampler for that; SimpleSampler is for getting a known-good starting point fast, not for fine control.