⌫ CLIP Negative Encode (Illustrious)

Negative prompts genuinely do something on Illustrious - it runs real classifier-free guidance, unlike a lot of the distilled 2026-era models where the negative pass isn't even computed. That means a good negative preset here is doing real work, not just a superstition carried over from older workflows. This node's whole purpose is giving you curated negative tag lists instead of you maintaining and pasting your own.

Preset vs. custom

preset defaults to illustrious_optimized and offers basic, comprehensive, ultra_clean, and custom alongside it. The general community wisdom on Illustrious negatives is that worst quality, low quality is usually sufficient and a forty-keyword negative is mostly cargo - that's roughly the spread this dropdown exposes as a tradeoff: ultra_clean sits closer to minimal, comprehensive is the long list, and illustrious_optimized is the pack's own tuned middle ground. Only custom hands full control back to your typed text field - with the built-in presets, expect the pack's own curated list to be doing the actual work rather than what you typed in the box.

strength_multiplier (0.5–2, default 1) scales the whole negative up or down without retyping anything - turn it down if a preset is fighting your positive prompt harder than you want, up if it's not doing enough. This is a faster lever than manually reweighting individual negative tags.

clip_skip here works exactly like it does on the positive encoder: 2 is the correct anime-lineage convention, and on SDXL architecture both text encoders already default to conditioning near that layer, so the setting mostly matters as "don't set it to 1," not as a fine-tuning dial.

The inputs and outputs that matter

• clip and text - required; text ships with a starter negative (worst_quality, bad_quality, lowres) so the node isn't empty by default.
• preset - the field that actually decides most of what gets encoded, unless set to custom.
• strength_multiplier - the fast way to tune a preset's overall bite without hand-editing tags.

Output is a single CONDITIONING - wire it to your sampler's negative input.

Installing it

ComfyUI Manager: search "EasyIllustrious," install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/regiellis/ComfyUI-EasyIllustrious
pip install -r ComfyUI-EasyIllustrious/requirements.txt

Nothing else to download - this runs on your checkpoint's own CLIP.

Common issues

Not sure whether your typed text is being used or ignored. This is worth testing directly on your own setup: switch preset to custom to guarantee your literal text is what gets encoded, and compare against a non-custom preset to see the difference. The schema doesn't specify whether presets append to or replace your typed text, so don't assume either way without checking.

Negative feels like it's fighting the prompt too hard, or an over-aggressive style is bleeding through. Drop strength_multiplier before switching presets entirely - it's the quicker adjustment and doesn't lose whatever else the preset is doing correctly.

Output still oversaturated despite a strong negative preset. A negative prompt can only push back so hard against CFG that's already too high - check cfg on your sampler first. Illustrious's comfortable range is roughly 4–7, and no negative preset fully compensates for CFG sitting well above that.