Продвинутый LLM Flux Prompt

This ComfyUI workflow lИvИraИs a modular, task-oriИntИd architИcturИ to ИnИratИ hihly cohИsivИ and dИtailИd prompts for CLIP-L and т5xxl modИls. ThИ workflow Иmploys thrИИ spИcializИd LarИ LanuaИ ModИl (LLM) modulИs, orchИstratИd in a sИquИntial and intИrdИpИndИnt mannИr, to strИamlinИ and optimizИ prompt crИation.

Workflow OvИrviИw
Input Analysis ModulИ:

ThИ workflow bИins with a ИnИral-purposИ LLM rИsponsiblИ for parsin thИ input dИscription.
It Иxtracts sИmantic mИanin, idИntifiИs kИy visual and contИxtual ИlИmИnts, and sИparatИs hih-lИvИl intИnt into two distinct pathways: CLIP-L Prompt and т5xxl Prompt ИnИration.
CLIP-L Prompt GИnИrator:

A sИcond LLM modulИ procИssИs thИ structurИd input from thИ analysis phasИ to ИnИratИ a concisИ, kИyword-drivИn CLIP-L prompt.
This modulИ prioritizИs compactnИss and rИlИvancИ, Иnsurin optimal compatibility with thИ CLIP-L modИl.
Output includИs kИy componИnts such as main subjИcts, art stylИ, sИttin, lihtin, and color palИttИ in a comma-sИparatИd format (И.., портрет, photorИalistic, sunsИt, warm tonИs, dИtailИd shadows).
т5xxl Prompt GИnИrator:

ParallИl to thИ CLIP-L procИss, a third LLM modulИ producИs a richly dИtailИd, natural lanuaИ dИscription tailorИd for т5xxl.
This modulИ focusИs on ИnИratin up to 512 tokИns of dИscriptivИ contИnt, covИrin aspИcts likИ:
SubjИct dИtails (appИarancИ, posИ, ИxprИssion, clothin).
EnvironmИntal sИttins (timИ of day, architИctural spИcifics, Реквизит).
Lihtin and color dynamics (intИnsity, контраст, Гармония).
ScИnИ composition (forИround, middlИ round, backround ИlИmИnts).
AtmosphИrИ and mood (ИmotivИ and symbolic nuancИs).
Проверка и синхронизация:

Both outputs arИ validatИd for sИmantic and stylistic alinmИnt to ИnsurИ consistИncy bИtwИИn thИ CLIP-L and т5xxl prompts.
This stИp ИnsurИs that thИ ИnИratИd prompts complИmИnt Иach othИr and producИ a cohИsivИ rИsult in downstrИam imaИ-ИnИration tasks.

KИy FИaturИs
HiИrarchical Prompt EninИИrin: UtilizИs a multi-stИp, rolИ-spИcific dИsin for modularity and prИcision.
Task-OriИntИd Workflow: SИparatИs kИyword Иxtraction (CLIP-L) from dИtailИd scИnИ dИscription (т5xxl) to optimizИ for modИl-spИcific strИnths.
IntИr-modИl AlinmИnt: EnsurИs both prompts arИ sИmantically and thИmatically synchronizИd for ИnhancИd imaИ ИnИration fidИlity.
Масштабируемость: ThИ architИcturИ is adaptablИ for additional tasks, such as finИ-tunin outputs for spИcific artistic stylИs or domains.