ComfyUI Node
Semantic Variation
A ComfyUI node in Eric Prompt Enhancers with 15 inputs and 4 outputs.
Semantic Variation
- clip
- text
- llm_instructions
- base_text
- report
◄text__person__, __gestures__, beside __objects/**__ in __spaces/**__. __light/setups__.►
◄seed0►
◄variation1►
◄strength0.35►
◄far_jumps0►
◄far_strength0.90►
◄embedderauto►
◄avoid_recentfalse►
◄legacy_syntaxtrue►
◄extra_paths►
◄directionnone►
◄direction_amount0.0►
◄nsfwoff►
◄violence_filteroff►
CategoryEric Prompt Enhancers
Inputs (15)
| Name | Type | Default | Description |
|---|---|---|---|
| text | STRING | __person__, __gestures__, beside __objects/**__ in __spaces/**__. __light/setups__. | Prompt with wildcard syntax. Every __name__, tag query and {a|b|c} is a slot that can be varied. |
| seed | INT | 00–18446744073709550000 | Seed for the BASE picks. Keep it fixed and change `variation` to explore around one prompt. |
| variation | INT | 10–4294967295 | 0 = the base prompt untouched. 1, 2, 3... = different variations of it, each reproducible. |
| strength | FLOAT | 0.350–1 | How far every slot drifts within its list. 0.1 = nearest few, 0.5 = a quarter of the way out, 0.7 = halfway, 1.0 = far end. |
| far_jumps | INT | 00–16 | How many slots jump with far_strength instead of drifting. Which ones is decided by `variation`. |
| far_strength | FLOAT | 0.900–1 | Strength used for the slots that jump. |
| embedder | COMBO | auto | What 'similar' means. auto = connected clip, else Qwen3-VL-Embedding, else open_clip. qwen3vl: similar in meaning and look. openclip / graph_clip: similar in look. sentence: similar in meaning. lexical: spelling only (no model). |
| avoid_recent | BOOLEAN | false | Make several candidate variations and keep the one least similar to recently generated prompts. |
| clipopt | CLIP | Optional. A CLIP with a pooled output (CLIP-L/G). Used as the embedder when `embedder` is auto or graph_clip. | |
| legacy_syntaxopt | BOOLEAN | true | Also accept the v1 forms {animal} and {label:a|b|c}. |
| extra_pathsopt | STRING | Extra wildcard folders, one per line, highest priority. ComfyUI/models/wildcards is always searched. | |
| directionopt | COMBO | none | Lean every move along a named axis, e.g. ordinary_to_uncanny or new_to_decayed. Axes are example phrases in variability/directions.yaml; add your own in directions_user.yaml. strength still decides how far a slot may go; this decides which way. |
| direction_amountopt | FLOAT | 0.0-3–3 | 0 = no lean. Above 0 leans toward the axis's `to` side (uncanny, decayed, ...), below 0 toward its `from` side. 1.0 is a clear lean, 2-3 lets the axis dominate. The report shows each slot's lean so you can see whether an axis bites on a given list. |
| nsfwopt | COMBO | off | How far wildcard picks may go (a word match on each entry and on list names, no model). off = SFW: no suggestive, nude or explicit lists or entries suggestive = also lingerie, bikinis, sexy, seductive ... nudity = also nude, naked, topless ... explicit = everything, including lists under _nsfw/ Fix a wrong rating in _catalog_user.yaml (nsfw: on a list, ratings: on entries). Entries pairing minor-related and sexual terms are removed at every level (config.json "content_filter"). |
| violence_filteropt | COMBO | off | Hold back violent wildcard picks (a word match on each entry and on list names, no model). off = nothing held back no gore = blood and gore, torture, mutilation, corpses no gore or violent acts = also murder, execution, stabbing, strangling, suicide ... no violence or weapons = also guns, swords, knives, bombs, battle, war Only wildcard picks are filtered; text you type is left as is. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| text | STRING | — |
| llm_instructions | STRING | — |
| base_text | STRING | — |
| report | STRING | — |