ENGLISH

I designed 3qu3str14n 3st4t3 as a background LoRA concept for a quiet, high-end English country equestrian estate. I wanted heritage luxury with believable materials: old brick, aged timber, worn leather, polished brass, soft dawn mist and restrained cinematic light.

My 45-image dataset explores eleven spaces: the indoor arena, stable block, tack room, hay loft, wash bay, dawn paddock, clubhouse lounge, competition tack room, grooming area, veterinary room and outdoor arena. Each space has four compositions; a final ground-level stable aisle closes the set. Sand, brick pavers, parquet, grass, tile, rubber and linoleum give each location its own identity.

These are empty backgrounds, without people or horses, real brands, stable names, competition names or readable text. Saddles and equipment stay on racks, stands and hooks. The training dataset is prepared; training and model performance have not been verified.

Trigger: 3qu3str14n 3st4t3

I place the trigger at the beginning, followed by the location, camera angle, materials and lighting.

Example: 3qu3str14n 3st4t3, heritage tack room, diagonal view, saddle racks, waxed parquet, tan leather, warm wall sconces, cinematic photography

Example: 3qu3str14n 3st4t3, dawn paddock, ground-level view, dew-covered grass, wooden rail fences, soft golden mist

Suggested experiments only, not verified settings: after training an SDXL or Illustrious version on its chosen base, I would start a small test at 28 steps, CFG 5 and LoRA weight 0.6, then compare weights 0.4 and 0.8 with the same seed. These values are proposed trial points, not measured recommendations. I have not tested either base. Krea usage depends on the exact model and whether the workflow supports this trained LoRA architecture; there is no confirmed Krea compatibility or universal Krea CFG, step or weight setting.

arrssenne

FRANCAIS

J'ai concu 3qu3str14n 3st4t3 comme concept de LoRA de backgrounds pour un domaine equestre anglais haut de gamme, calme et feutre. Je voulais un luxe patrimonial credible : vieille brique, bois ancien, cuir patine, laiton poli, brume matinale et lumiere cinematographique discrete.

Mon dataset de 45 images explore onze espaces : manege couvert, bloc des ecuries, sellerie, grenier a foin, baie de lavage, paddock a l'aube, salon du clubhouse, sellerie de competition, aire de pansage, salle veterinaire et carriere exterieure. Chaque lieu compte quatre compositions; une derniere vue au sol de l'allee des ecuries ferme la serie. Sable, paves, parquet, herbe, carreaux, caoutchouc et linoleum donnent une identite propre a chaque espace.

Ce sont des backgrounds vides, sans personnes ni chevaux, marques reelles, noms d'ecuries ou de competitions, ni texte lisible. Le materiel reste sur ses supports et crochets. Le dataset est prepare; l'entrainement et les performances du modele ne sont pas verifies.

Trigger : 3qu3str14n 3st4t3

Je place le trigger au debut, suivi du lieu, du cadrage, des materiaux et de la lumiere. Les deux exemples anglais ci-dessus montrent cette structure.

Essais proposes seulement, sans validation : apres entrainement d'une version SDXL ou Illustrious sur sa base choisie, je commencerais un petit test a 28 steps, CFG 5 et poids LoRA 0.6, puis comparerais 0.4 et 0.8 avec le meme seed. Ces valeurs sont des points d'essai proposes, pas des recommandations mesurees. Aucune des deux bases n'a ete testee. L'utilisation dans Krea depend du modele exact et du support de l'architecture du LoRA entraine; aucune compatibilite Krea ni valeur universelle de CFG, steps ou poids n'est confirmee.

arrssenne