SliderObjectIn

The direction-finder nodes in this pack (SliderLatent and SliderLatentPooled) build their sample prompts from a random vocabulary of mediums - painting, photograph, sketch, that kind of thing - and subjects - dog, car, person, and so on. By default they use hardcoded lists of about 20 mediums and 70 subjects. SliderObjectIn is the valve that lets you swap those lists for your own.

It's the simplest node in the pack and the easiest to understand: two multiline text boxes, mediums and subjects, and two typed outputs, MEDIUMS and SUBJECTS, which only the SliderLatent nodes will accept. Type your words, connect the outputs to SliderLatent's optional mediums and subjects sockets, and the direction search now samples from your vocabulary instead of the defaults.

The gotcha that bites everyone

The node's splitting logic is one small function, and it has opinions. If the string contains a semicolon, it splits on ; and trims each item. Otherwise the whole trimmed string is treated as one single item - and an empty string becomes an empty list. So "man;woman;person" gives you three subjects, but typing the same words on separate lines gives you one subject that contains literal newlines. The multiline box is a red herring: semicolons are the separator. This is the classic mistake - someone newlines a list, wonders why the slider barely changes, and it's because they just handed the finder a subject called "man\nwoman\nperson."

One thing worth knowing: leaving a box empty is fine. SliderLatent checks whether it received anything and falls back to its built-in lists, so this node is fully optional. The author's own shipped workflow even includes it (mediums "photograph", subjects "man;woman;person") but never wires its outputs to anything - which is about as strong a statement as a workflow can make that it's a convenience, not a requirement.

When you'd actually reach for it

• Your direction keeps drifting toward irrelevant subjects, so you narrow the pool: only "man;woman;person", only "photograph".
• You're chasing a niche concept (say, "cyberpunk" vs "rustic") and want the sampling to stay in a subject domain where that distinction is legible.
• You want a bigger or different vocabulary than the built-ins - dump in your own comma... no, semicolon-separated list.

Because it only reshapes the sampling distribution, it changes which prompts the direction averages over, not the mechanism. Same install story as the rest of the pack: ComfyUI Manager → search "ComfyUI-CLIPSlider-SC", or git clone https://github.com/chenbaiyujason/ComfyUI-CLIPSlider-SC into custom_nodes, then restart. No models, no heavy deps, and - since the README is inherited from the original RhizoNymph project and never mentions this node - the shipped workflow is your best guide to how it fits.