◓ Poses

Worth clearing up before anything else: this node doesn't touch an image, a skeleton, or a ControlNet map. The pack's README lists it as "pose reference integration with OpenPose compatibility," which is easy to misread as "feed it an OpenPose skeleton." What it actually does is inject a text tag describing a pose into your prompt - the "compatibility" is that these tags describe poses in vocabulary that overlaps with what an actual OpenPose ControlNet would produce, not that the node reads pose images. Check the node's sockets and there's no IMAGE or POSE_KEYPOINT input anywhere - just enums and strings.

How it works

Pose Set is the core pick - 44 choices, all in the ALL_CAPS style Danbooru-adjacent tools tend to use internally: CROUCHING, SQUATTING, WIDE_STANCE, ONE_KNEE_UP, HAND_BETWEEN_LEGS, and so on. Format decides whether the output keeps that KEY form or renders it as ordinary words instead - leave it off for readable output, flip it on if you specifically want the raw key.

Descriptive (BETA) is the more interesting toggle: instead of inserting the single pose tag, it switches to a natural-language description of the pose. This leans on something specific to Illustrious's training - it handles plain sentences noticeably better than most SDXL-era anime models, though tags remain the model's primary, most reliable interface. Since this mode is explicitly marked Beta in the tooltip, treat it as the experimental option: try it, but don't be surprised if the tag form gives you a more predictable result.

Pose Weight (0.1–2, default 1) applies the standard (tag:weight) emphasis syntax to whatever pose text gets inserted. Offset (Beta) is the other flagged-experimental field - it controls how many words back from the end of your prefix text the pose gets spliced in, rather than always appending it at the very end. Also Beta, also worth expecting some rough edges.

Add Chain Insert Point (optional) drops the pack's [EN122112_CHAIN] placeholder so a downstream node can inject its own content at that exact spot - the same chaining mechanism used across the suite's generator nodes. prefix and suffix (optional strings) round out the usual plumbing.

Output is a single PROMPT string.

Installing it

Via ComfyUI Manager: search "EasyIllustrious," install, restart - the recommended route. Manually: cd ComfyUI/custom_nodes && git clone https://github.com/regiellis/ComfyUI-EasyIllustrious, then pip install -r ComfyUI-EasyIllustrious/requirements.txt inside your ComfyUI environment, then restart.

Where the rough edges actually are

Both of the node's more powerful features - Descriptive mode and Offset - carry an explicit (Beta) label in their own tooltips, which is the pack telling you directly where the edge cases live. If the pose lands in an odd spot in your final prompt, or Descriptive mode's phrasing reads awkwardly next to your other tags, that's the acknowledged experimental behavior, not a bug worth chasing down. The safe fallback is Offset at 0 (append to the end, no repositioning) with Descriptive off (plain tag form) - get a reliable base result there first, then experiment with the Beta features once you know what the plain version looks like.

If you actually need precise, image-derived pose control rather than a tag-level description, that's a job for an OpenPose ControlNet feeding your sampler, not this node - the two are complementary, not interchangeable.