LoRA Suggester

A ComfyUI node in Eric Prompt Enhancers with 25 inputs and 5 outputs.

By EricRollei·Created a day ago·Updated about 17 hours ago· 3

LoRA Suggester

sections

suggestions
triggers
prompt_with_triggers
report
lora_text

◄platformkrea2►

◄modesuggest►

◄nsfwoff►

◄max_loras4►

◄style_max2►

◄treatment_max2►

◄detailer_max1►

◄environment_max1►

◄lighting_max1►

◄subject_max1►

◄layout_max1►

◄utility_max1►

◄apply_threshold0.35►

◄weight_budget2.5►

◄stage_profilerole►

◄variety0.20►

◄seed0►

◄text►

◄exclude►

◄pin►

◄transform_max1►

◄slider_max2►

◄slider_range3.0►

◄slider_min_strength2►

CategoryEric Prompt Enhancers

Inputs (25)

Name	Type	Default	Description
platform	COMBO	krea2	Target model: only LoRAs for it are considered (its LoRA shelf folder / Civitai base model).
mode	COMBO	suggest	suggest: report only, nothing loads. apply: send the picks to the loader.
nsfw	COMBO	off	Highest NSFW level a LoRA may have. NSFW LoRAs are used only when the prompt itself asks for that content. off still allows suggestive styles, and styles that can paint nudes (turned down a little).
max_loras	INT	40–10	Most LoRAs in the stack, all roles together.
style_max	INT	20–4	Most style LoRAs (base look: artist, movement, illustration or photo style). 0 = never.
treatment_max	INT	20–4	Most treatment LoRAs (layered on a style: linework, halftone, paper, grain, colour grade, material looks). 0 = never.
detailer_max	INT	10–4	Most detailer LoRAs (detail, sharpness, skin, hands, anatomy). 0 = never.
environment_max	INT	10–4	Most environment LoRAs (location or background). 0 = never.
lighting_max	INT	10–4	Most lighting LoRAs (light and colour mood). 0 = never.
subject_max	INT	10–4	Most subject LoRAs (character, person, creature, object, clothing, pose). 0 = never.
layout_max	INT	10–4	Most layout LoRAs (composition, page layout, framing). 0 = never.
utility_max	INT	10–4	Most utility LoRAs (speed LoRAs, nudity enablers - only when clearly asked for). 0 = never.
apply_threshold	FLOAT	0.350–1	How good a match must be before a LoRA is used. Lower = adventurous stacks, higher = only clear matches.
weight_budget	FLOAT	2.50–10	Most total Stage-1 weight. The strongest pick keeps its weight, the others are scaled down to fit. 0 = no limit.
stage_profile	COMBO	role	role: each role gets its own S1/S2/S3 shape (detailers late, environments early). flat: the same weight in every stage.
variety	FLOAT	0.200–1	0 = always the best match. Higher = with a new seed, close runners-up can win instead.
seed	INT	00–2147483647	Seed for variety. Connect the Composer's seed to re-roll LoRAs with the prompt.
sectionsopt	PROMPT_SECTIONS		From a Wildcard Composer: each section feeds its roles (style -> style and treatment, setting -> environment, ...).
textopt	STRING		Plain prompt text, when there are no sections (or in addition to them).
excludeopt	STRING		Never use these LoRAs. One entry per line or comma separated; case does not matter. Whole words only: 'age' matches 'Age Slider' but not 'image' or 'vintage'; 'hair' does not match 'chair'. Several words must appear together: 'waist slider'. plain words - the LoRA's name or file name (vallejo, hair, waist slider) folder:<words> - its folder below the LoRA root (folder:sliders, folder:art-style) role:<role> - a Suggester role (role:treatment, role:detailer, role:slider) axis:<axis> - sliders on that axes.yaml axis (axis:age, axis:body weight) The report shows how many LoRAs each entry excluded.
pinopt	STRING		Always use these LoRAs, whatever their score (they skip the threshold, role limits and NSFW gate, and count toward max_loras). One entry per line or comma separated; whole words. a name - ONE LoRA: the best match (your stars, favourite, card confidence). If the words match several, the report lists the others - add more of the name to pick a different one. name=weight - that LoRA at this weight (all stages; a slider keeps its stage shape): 'waist slider=-8', 'vallejo=0.7' axis:<axis>=weight - the best slider on that axis at this weight: 'axis:age=+3' Give sliders a weight; without one they run at their card's typical weight. folder: and role: work in exclude only.
transform_maxopt	INT	10–4	Most transform LoRAs (changes content and composition but stacks on a style: surreal, dreamlike, glitch). 0 = never.
slider_maxopt	INT	20–6	Most slider LoRAs (dials (age, body, shot size, saturation ...), set from what the prompt asks for on that axis). They do not count toward max_loras or the weight budget. 0 = never.
slider_rangeopt	FLOAT	3.00.1–15	For a slider whose card has no believable range: it dials within -range..+range (a strongest-cue prompt gets the full range). Ranges set in the card editor, or found by the slider card pass, are used instead.
slider_min_strengthopt	INT	21–3	How clearly the prompt must ask before a slider is used, by the cue's position on its axis (axes.yaml ladder, -1..1). 1 = any cue; 2 = 0.35 or more (everyday words stay off); 3 = 0.65 or more, only strong asks ('wasp waist', 'ancient').

Outputs (5)

Name	Type	Description
suggestions	LORA_SUGGESTIONS	The picked LoRAs with stage weights. Connect to the Eric Krea2 Multi-LoRA Stack's `suggestions` input. Empty in suggest mode.
triggers	STRING	Trigger words of the picks, comma separated.
prompt_with_triggers	STRING	The prompt with the triggers added (for workflows without a rewriter; with a rewriter, let the Multi-LoRA's add_triggers merge them after the rewrite instead).
report	STRING	What was picked and why, the runner-up per role, and what was excluded.
lora_text	STRING	The picks as <lora:name:weight> tags, one weight per LoRA (the strongest of its three stage weights), for loaders outside Eric_Krea2 that read this format (e.g. LoRA Manager's text loader). Empty in suggest mode.