ENGLISHr0c0c0 m4n0r - Rococo Mansion BackgroundsI designed this background LoRA concept to recreate the interiors and garden of an imaginary 18th-century French Rococo aristocratic mansion. I wanted its charm to come from the decor alone: powder pink, celadon blue and ivory, antique gold carving, silk and damask, aged mirrors, crystal chandeliers and warm candlelight.I explored eleven distinct locations: the grand reception salon, intimate boudoir, canopy bedroom, gala dining room, cabinet of curiosities and library, music salon, marble bathing room, hall of mirrors, orangery, private theatre and formal garden at dusk. A final enfilade connects the salon, mirror gallery and twilight garden.I prepared 45 square 1024x1024 training images, with matching captions and individual prompts. The backgrounds are empty, with furniture kept toward the perimeter where the room composition allows an open space for a character. The compositions include wide views, low immersive angles, elevated views and diagonals through furniture or architecture. The mansion is a creative Rococo interpretation; the glowing boudoir vanity and other requested details are not a claim of strict historical reconstruction.CURRENT STATUSThis package is a prepared training dataset and a draft release description. I have not established a trained LoRA release, tested checkpoint compatibility or validated inference settings for this concept.TRIGGERI use the exact trigger r0c0c0 m4n0r at the beginning of the prompt, followed by the room, viewpoint, light and desired atmosphere. The trigger does not load a LoRA by itself: I also need to select the matching trained LoRA in my generation workflow.SUGGESTED STARTING POINTS - NOT VERIFIED RESULTSFor a future SDXL-trained version, my proposed initial test is 1024x1024, 30 steps, CFG 5 and LoRA weight 0.7. I would compare 25-35 steps, CFG 4-6 and weights 0.5-0.9 on fixed seeds rather than presenting any value as optimal.For an Illustrious-trained version, I would begin with the same small test grid on the exact training base or its intended compatible checkpoint. This is a proposed experiment, not demonstrated compatibility.For Krea, I would first identify the actual model and use a LoRA trained for that model. I do not assume that an SDXL or Illustrious LoRA transfers to Krea. If I separately train a Krea 2 version on RAW, the official Krea 2 repository provides RAW examples at 52 steps / CFG 3.5 and Turbo examples at 8 steps with CFG disabled (0.0 in its reference implementation). Those are model workflow references, not settings tested with r0c0c0 m4n0r. I would initially test a weight of 0.7 only if the interface exposes that control; otherwise I would keep its default strength and record the actual value.No SDXL, Illustrious or Krea compatibility is confirmed for this dataset, and no trained model is included here.PROMPT STRUCTURES TO TRYr0c0c0 m4n0r, grand reception salon, wide establishing shot, gilded boiseries, pastel silk, crystal chandeliers, open center floor, soft daylight, no humansr0c0c0 m4n0r, intimate boudoir, diagonal view through a botanical dressing screen, pink velvet chaise longue, aged mirrors, warm vanity glow, no humansr0c0c0 m4n0r, formal garden at dusk, low angle along a gravel alley, stone fountain, conical topiary, blue twilight, no humansFor a character scene, I would keep the room and viewpoint, replace the no humans instruction with an explicit adult character description, and test the result after training. Character integration has not been validated in this package.Model reference for the Krea 2 examples: <a target="_blank" rel="ugc" href="https://github.com/krea-ai/krea-2/blob/main/README.md">https://github.com/krea-ai/krea-2/blob/main/README.md</a>The SDXL/Illustrious test ranges and LoRA weights above are my proposed starting experiments, not results taken from a trained release.arrssenneFRANCAIS SANS ACCENTSr0c0c0 m4n0r - Backgrounds de manoir RococoJ'ai concu ce concept de LoRA de backgrounds pour recreer les interieurs et le jardin d'un manoir aristocratique francais imaginaire du XVIIIe siecle, dans un esprit Rococo. Je voulais que son charme vienne du decor seul : rose poudre, bleu celadon et ivoire, sculptures dorees, soie et damas, miroirs anciens, lustres de cristal et lumiere chaude des bougies.J'ai explore onze lieux distincts : grand salon de reception, boudoir intime, chambre a baldaquin, salle a manger de gala, cabinet de curiosites et bibliotheque, salon de musique, salle de bain en marbre, galerie des miroirs, orangerie, theatre prive et jardin formel au crepuscule. Une derniere vue en enfilade relie le salon, la galerie et le jardin.J'ai prepare 45 images carrees de 1024x1024, avec leurs tags et leurs prompts individuels. Les backgrounds sont vides, avec des meubles en peripherie lorsque la composition permet de laisser une place pour un personnage. Les cadrages comprennent des vues larges, des angles bas immersifs, des plongees et des diagonales a travers le mobilier ou l'architecture. Le manoir est une interpretation creative du Rococo; la coiffeuse lumineuse du boudoir et certains details demandes ne constituent pas une reconstruction historique stricte.ETAT ACTUELCe dossier contient un dataset prepare pour l'entrainement et une description de publication a valider. Je n'ai pas etabli de version entrainee du LoRA, teste sa compatibilite avec des checkpoints ou valide ses reglages de generation.TRIGGERJ'utilise le trigger exact r0c0c0 m4n0r au debut du prompt, puis le lieu, le cadrage, la lumiere et l'ambiance. Le trigger seul ne charge pas un LoRA : je dois aussi selectionner le LoRA entraine correspondant dans mon workflow.POINTS DE DEPART SUGGERES - RESULTATS NON VERIFIESPour une future version entrainee sur SDXL, mon premier essai propose est 1024x1024, 30 steps, CFG 5 et poids LoRA 0.7. Je comparerais 25-35 steps, CFG 4-6 et des poids de 0.5-0.9 sur des seeds fixes, sans declarer une valeur optimale.Pour une version entrainee sur Illustrious, je commencerais par la meme petite grille de tests sur la base exacte d'entrainement ou le checkpoint compatible vise. Il s'agit d'un essai propose, pas d'une compatibilite demontree.Pour Krea, j'identifierais d'abord le modele reel et j'utiliserais un LoRA entraine pour ce modele. Je ne presume pas qu'un LoRA SDXL ou Illustrious se transfere a Krea. Si j'entraine separement une version Krea 2 sur RAW, le depot officiel donne des exemples RAW a 52 steps / CFG 3.5 et Turbo a 8 steps avec CFG desactive (0.0 dans son implementation de reference). Ce sont des references de workflow du modele, pas des reglages testes avec r0c0c0 m4n0r. Je testerais un poids de 0.7 uniquement si l'interface expose ce controle; sinon je garderais sa force par defaut et consignerais la valeur reelle.Aucune compatibilite SDXL, Illustrious ou Krea n'est confirmee pour ce dataset, et aucun modele entraine n'est inclus.STRUCTURES DE PROMPTS A ESSAYERr0c0c0 m4n0r, grand reception salon, wide establishing shot, gilded boiseries, pastel silk, crystal chandeliers, open center floor, soft daylight, no humansr0c0c0 m4n0r, intimate boudoir, diagonal view through a botanical dressing screen, pink velvet chaise longue, aged mirrors, warm vanity glow, no humansr0c0c0 m4n0r, formal garden at dusk, low angle along a gravel alley, stone fountain, conical topiary, blue twilight, no humansPour une scene avec personnage, je garderais le lieu et le cadrage, remplacerais no humans par la description explicite d'un personnage adulte, puis testerais le resultat apres entrainement. L'integration d'un personnage n'a pas ete validee dans ce dossier.Reference de modele pour les exemples Krea 2 : <a target="_blank" rel="ugc" href="https://github.com/krea-ai/krea-2/blob/main/README.md">https://github.com/krea-ai/krea-2/blob/main/README.md</a>Les plages de tests SDXL/Illustrious et les poids LoRA sont mes essais de depart proposes, pas des resultats issus d'une version entrainee.arrssenne

Files with SHA256: 082a5681a41f3d9ea372d884b469c0a82b29e7fecf94eaf48a6412c27249e7ac