ENGLISH
I created c4b4r3t n1ght to recreate the atmosphere of a grand 1930s cabaret after dark, with glamour carried entirely by its decor: deep red velvet, carved dark wood, brass, gilt ornament, warm bulbs and cinematic haze.
The set explores eleven spaces: the stage, private booths, curved bar, performers' dressing rooms, champagne lounge, wardrobe and makeup annex, lighting booth, gilt mezzanine, VIP smoking room, abstract poster corridor and rain-soaked entrance. A final elevated view brings the venue together. The dataset contains 45 square background images with varied wide, low, elevated and diagonal compositions.
These are backgrounds only, with no people or worn costumes. The venue is fictional and generic; readable text, real-world branding and venue names are excluded. Posters and the neon marquee use abstract shapes. Costumes hang on racks.
Use the exact trigger: c4b4r3t n1ght
Place it near the beginning of the prompt, then describe the room, camera angle, materials and its specific lighting.
Prompt structures:
c4b4r3t n1ght, grand 1930s cabaret stage, wide view, red velvet curtains, worn wooden boards, single amber spotlight, cinematic haze
c4b4r3t n1ght, private cabaret booth, diagonal view through a crimson velvet curtain, tufted seating, brass table lamp, patterned carpet
c4b4r3t n1ght, cabaret entrance at night in the rain, abstract neon marquee, brass gates, wet pavement reflections
Suggested starting points only — not tested results:
For a future SDXL-family or Illustrious version trained on the corresponding base, I would begin exploration around 25–35 steps, CFG 4–7 and LoRA weight 0.6–0.9, then adjust one variable at a time to suit the checkpoint and sampler. These ranges are provisional suggestions; no training, generation test or compatibility validation has been performed for this LoRA.
SDXL and Illustrious are potential training targets, not verified compatible versions of a single file. Krea is also untested: use these scene descriptions only if the selected workflow supports them, and confirm whether that workflow supports the trained LoRA. I am not claiming Krea LoRA import support or proposing verified Krea steps, CFG or weight controls.
This delivery is the training dataset, not a trained model. Final behavior, compatibility and settings remain to be established after training.
arrssenne
FRANCAIS
J'ai cree c4b4r3t n1ght pour retrouver l'atmosphere d'un grand cabaret des annees 1930 a la nuit tombee. Le glamour vient entierement du decor : velours rouge profond, bois sombre sculpte, laiton, dorures, ampoules chaudes et brume cinematographique.
Le set explore onze espaces : scene, alcoves privees, bar courbe, loges, salon champagne, annexe costumes et maquillage, cabine d'eclairage, mezzanine doree, fumoir VIP, corridor d'affiches abstraites et entree sous la pluie. Une derniere vue plongeante rassemble la salle. Le dataset contient 45 backgrounds carres avec des compositions larges, basses, plongeantes et diagonales variees.
Ce sont uniquement des decors, sans personnes ni costumes portes. Le lieu est fictif et generique; textes lisibles, marques reelles et noms d'etablissement sont exclus. Les affiches et la marquise utilisent des formes abstraites. Les costumes restent sur les portants.
Trigger exact : c4b4r3t n1ght
Je le place au debut du prompt, puis je decris la piece, l'angle de camera, les materiaux et son eclairage particulier. Les trois exemples anglais ci-dessus montrent cette structure.
Points de depart suggeres seulement — aucun resultat teste :
Pour une future version SDXL ou Illustrious entrainee sur la base correspondante, je commencerais les essais autour de 25–35 steps, CFG 4–7 et poids LoRA 0.6–0.9, puis j'ajusterais une variable a la fois selon le checkpoint et le sampler. Ces plages sont provisoires; aucun entrainement, test de generation ou controle de compatibilite de ce LoRA n'a ete effectue.
SDXL et Illustrious sont des cibles d'entrainement possibles, pas des versions compatibles deja verifiees d'un fichier unique. Krea reste non teste : utiliser les descriptions de scene si le workflow choisi les accepte, puis verifier sa prise en charge du LoRA entraine. Je ne confirme pas l'import de LoRA dans Krea et je ne presente aucun controle steps, CFG ou poids Krea comme verifie.
Cette livraison est le dataset d'entrainement, pas un modele entraine. Le comportement final, les compatibilites et les reglages devront etre etablis apres entrainement.
arrssenne