ꘅ Tutorial

What it is

ꘅ Tutorial is a bookmark wearing a node costume. It has no inputs, no outputs, and doesn't run anything. It draws a single ▤ button that opens the pack's Tutorial 0 page in a new browser tab. That's the whole feature.

So why does it deserve a page? Because this pack has an unusual problem for a ComfyUI add-on: it isn't really a node collection, it's an argument about how AI decisions should be wired into creative work. Tutorial 0 is where that argument is made, offline, with no model and no download. If you install anything from genereticore, open this first. It's ten minutes and it will save you from building a graph around a premise you misunderstood.

What the tutorial actually teaches

The pattern. You author a state (a description of a scene), a question, and a fixed set of allowed labels. A local model ranks those labels. A mapping node of your own turns the winning label into a number. The model never draws, never writes code, never invents an option - it makes one bounded choice inside a design you built.

The contract between parts. Labels are strings you define. blue and orange are not colors to Comfy; they're answer names. They become a hue only when ꘅ Choice to Value maps them to 210 and 30. The tutorial is explicit that adding a third label means adding a third mapping, because a label alone tells nothing downstream what to render. That's the single most common way a graph from this pack gets built wrong.

Why not just prompt an LLM? This is the part I'd point a sceptic at. A chat model generates a sequence of tokens; a specialist like d1 evaluates your allowed answers in one forward pass and emits nothing else. Those are different operations with different failure modes, and the second one can't drift off-format. The general version of that argument - messy prose in, regex out - is the reason most LLM-in-graph setups need a cleaner bolted to the output. Here the answer is the format.

Scores are not confidence. The tutorial walks through three distributions the pack actually emits, each tagged with its own method: d1_temperature_scaled for the specialist, answer_slot_softmax_uncalibrated for the Qwen answer-slot path, normalized_uncalibrated_scores for supplied scores. Then it says the thing almost nobody writing LLM nodes says out loud: this does not mean 80 percent of people would prefer blue, and it doesn't mean blue is correct. A model can be confidently unhelpful. Two nearly equal scores mean "don't switch palettes yet."

Using it

cd ComfyUI/custom_nodes
git clone https://github.com/languel/genereticore

Restart Comfy, or install via ComfyUI Manager by searching GeneretiCore. The node appears under Genereti / Inference as ꘅ Tutorial.

In a graph, it's a labelled sticky note with a live link - drop it next to the decision nodes in a shared workflow so whoever opens the file next can read the theory without you explaining it. Pressing Run just returns a hint string in its UI payload; nothing executes, no download, no model.

If you'd rather read it on disk, the lesson is in the repo at docs/tutorial-0.md, next to docs/getting-started.md (which has the button-by-button table and the troubleshooting list). The best entry point, though, is the sample: Workflows → GeneretiCore → 00 - Tutorial 0 - decision models. It installs itself into your Comfy profile on first startup, and it opens with the book button pressed.

Where people get burned

The samples are in Workflows, not the node menu. On a first install, if you don't see 00 - Tutorial 0 in the Workflows panel, restart Comfy and then refresh the panel. The pack installs example_workflows/ into the profile on startup and preserves edited copies; replaced or retired unchanged samples get backed up to .genereticore-workflow-backups. You can also drag any JSON out of example_workflows/ straight onto the canvas.

Samples 05, 06, 07 want a second pack. Those three need the separate public Genereti package installed into the same Comfy instance, plus its p5 editor and walkthrough player. You'll see red missing nodes otherwise. Samples 00–04, including this tutorial and the no-model decision sample, need nothing but GeneretiCore and stock Comfy nodes.

No model, no problem. Everything this node points at is readable and runnable without any Core AI helper or platform gate - which makes it the one part of this macOS-27-bound pack that's useful to literally everyone, including people who just want to understand the pattern before deciding whether the rest is worth the install.