✦ ALCHEMIST // LORA CAPTION

Write one training caption. Describes what should stay promptable and stays silent about whatever must bind to the trigger. lora_kind picks the rule.

By babydjac·Created 5 days ago·Updated 5 days ago· 0

◄actionidle►

◄providerxAI / Grok►

◄modelgrok-4.20-non-reasoning►

◄reasoningprovider default►

◄lora_kindbody concept►

◄trigger_word►

◄nsfwtrue►

◄simple_prompt►

◄caption_words45►

◄omit_terms►

◄extra_direction►

◄api_key►

◄temperature0.80►

◄max_tokens1400►

◄remember_keytrue►

◄custom_base_url►

◄reroll0►

Categoryprompting/alchemist

Inputs (18)

Name	Type	Default	Description
action	COMBO	idle	Runs in the browser and resets to idle. Never queues a generation.
provider	COMBO	xAI / Grok	Which LLM answers. Use FETCH MODELS after switching.
model	COMBO	grok-4.20-non-reasoning	Model id. FETCH MODELS repopulates this from the live API.
reasoning	COMBO	provider default	Only bites on reasoning models, where it is the difference between 9s and 42s. Ignored elsewhere.
lora_kind	COMBO	body concept	Decides what is omitted so it binds to the trigger.
trigger_word	STRING		Placed first in every caption.
nsfw	BOOLEAN	true	—
source_imageopt	IMAGE		—
simple_promptopt	STRING		—
caption_wordsopt	INT	4520–70	—
omit_termsopt	STRING		—
extra_directionopt	STRING		—
api_keyopt	STRING		Leave empty to use the vaulted key or the provider env var.
temperatureopt	FLOAT	0.800–2	—
max_tokensopt	INT	140064–32000	—
remember_keyopt	BOOLEAN	true	—
custom_base_urlopt	STRING		Override the provider endpoint. Required for the Custom provider.
rerollopt	INT	00–4294967295	Output is cached while inputs are unchanged, so repeat queues are free. Bump this to force a fresh call.