excel_roles

Describing a character in prompt form usually means juggling a dozen small details at once - hairstyle, face shape, eye shape, build, what they're wearing, what they're doing, how they're feeling - and keeping all of that consistent in your head while typing a paragraph. excel_roles turns that into a set of individual dropdowns instead: one for each attribute, so you build the description piece by piece rather than composing it from scratch every time.

The attributes on offer, and roughly how granular each one is: role (32 options - character archetypes like princess, dancer, athlete team captain), clothing (288, by far the biggest list - everything from casual daily wear to formal shirts), hairstyle (32), skin (10 - skin tone and finish, including things like tan lines), face_shape (13), ears (12), eyes (21), eyebrows (21), nose (11), mouth (10), figure (12 - body build), chest (20), action (101, the second-biggest list - running, skateboarding, ballet, martial arts poses, and so on), and expression (17 - angry, surprised, wistful, and the rest). There's also a template field with 73 pre-built named character combinations if you'd rather start from a curated character and tweak specific fields than assemble one from thirteen separate dropdowns.

Like excel_VedioPrompt, this node does actual template substitution rather than plain concatenation, and its own description gives a worked example: "{role}显得很{expression}" (roughly, "{role} looks very {expression}") with role set to "警察" (police officer) and expression set to "惊讶" (surprised) produces "警察显得很惊讶" - police officer looks very surprised. So whichever fields you set feed into the template's placeholders by name.

As with the rest of the excel_* family, the dropdown content is predominantly in Chinese - this is a China-based author's pack, and this node hasn't been translated tag-by-tag. There's a required language toggle (default on), which is presumably a translation switch for the output, but the dropdown labels themselves will still read in Chinese regardless of that setting. There's also an optional multiline custom field if you'd rather write your own template structure instead of relying on the built-in ones.

Output. A single STRING, pos - the assembled character description, ready to feed into a text encoder or wherever else in your workflow expects a prompt.

Where this actually helps. If you're generating a set of images meant to feature the same character repeatedly - a comic, a set of product shots with a consistent model, concept art variations - having each physical attribute as its own dropdown makes it much easier to change one thing (swap the expression, change the action) while keeping everything else identical, compared to hand-editing a paragraph and hoping you didn't accidentally alter a detail you meant to keep.

Installing it. Standard for the whole pack - search "ComfyUI-Apt_Preset" in ComfyUI Manager, or clone directly:

cd ComfyUI/custom_nodes
git clone https://github.com/cardenluo/ComfyUI-Apt_Preset.git

Run install.bat on Windows for dependencies, then restart. Pure text assembly, no models or GPU work - it should function the moment the pack loads.

Troubleshooting. If a placeholder in a custom template isn't substituting, check the exact key name against the fields this node exposes ({role}, {clothing}, {hairstyle}, {skin}, {faceshape}, {ears}, {eyes}, {eyebrows}, {nose}, {mouth}, {figure}, {chest}, {action}, {expression}, per its own description) - an unrecognized key is left as literal text rather than raising an error. If the whole pack fails to load, that's the documented IMPORT FAILED issue from a missing dependency elsewhere in the install, not this node specifically - check your ComfyUI console log on startup and reinstall via Manager's "install missing custom nodes."