DDRK Omega Smart Config

Somewhere between "which sampler should I use" and "why is my Flux workflow using an SDXL scheduler" there's a simpler question people rarely ask out loud: what model is this, actually? DDRK Omega Smart Config exists to answer that. Feed it a model and a latent_image and it inspects the checkpoint, tells you which family it belongs to, and hands back a full recommended parameter set as node outputs. It's the pack's introspection node, and it's refreshingly honest about what it won't do.

What it figures out

Under the hood it reads the model's unet config - the image_model string, adm_in_channels, context_dim - plus the sigma range and the latent's channel count, and classifies the checkpoint: flux, sdxl, sd15, sd2, sd3, qwen, krea, hidream, chroma, lumina, or a generic fm/edm fallback. It prints the detected family to your console ([DDRK SmartConfig] Detected family: ...) so you can see it even without reading the output wires.

The outputs, and how to use them

• family (STRING) - the detected architecture family. The one you'll actually read.
• hint (STRING) - a human-readable suggestion, e.g. that sharpening is disabled on EDM, or to use a distilled LoRA for 4-8 steps.
• guidance_embed (BOOLEAN) - whether the model expects a guidance embedding, useful when you're wiring up the conditioning side.
• scheduler_type, flow_shift, integrator, sde_strength, sharpness, saber_fusion, momentum_beta - a recommended starting point for each of the parameters the pack's samplers expose.

The clean way to use it: wire those STRING/FLOAT outputs directly into the matching inputs on the DDRK Omega Unified KSampler (or the Sampler + Scheduler pair). Then your sampler and scheduler are never out of sync with what the node detected, and when you swap checkpoints the config re-derives itself. It's read-only - it samples nothing, changes nothing.

What it deliberately doesn't do

It will not guess your steps or CFG. The tooltip and the README are both explicit: those depend on training, LoRA and distillation, so they're checkpoint-specific and stay yours to set. That restraint is the whole design philosophy of this pack - the author's stated goal is a node you drop in, not one you tune for an hour, and a config node that started inventing CFG values would be exactly the kind of magic that bites you later.

Installing it

Same pack, same zero-extra-dependency install:

cd ComfyUI/custom_nodes/
git clone https://github.com/HVOSTOVSKY/DDRK-Omega-Sampler.git

Restart ComfyUI, or install via ComfyUI Manager by searching "DDRK Omega Sampler".

Gotchas

• On exotic or unusual checkpoints the family can come back unknown, in which case it falls back to a coarse EDM-vs-FM split from the sigma range. The recommended values in that case are starting points, not tuned optima - the README says as much for everything here.
• The recommendations are per-family defaults, and the author explicitly notes the enhancer levels (SABER, sharpening) are "on by reputation, not by measurement." Treat the outputs as a sane base, not a promise.
• Don't expect this to set up your whole workflow. It sets the sampler-adjacent knobs and nothing else, and that's the point.