ENGLISHWith st0n34g3 sub0rb, I wanted to make everyday suburban life feel as if it had been built from enormous slabs of stone, weathered logs, woven fibers, fur and shells. I imagined a cheerful prehistoric town inside a red-rock canyon, with rounded houses, warm windows, lush cycads and a smoking volcano on the horizon. The mood is deliberately theatrical: imposing furniture, dramatic sunlight, deep shadows and a little handmade anachronistic humor.I built this background LoRA dataset around eleven distinct spaces: living room, kitchen, bedroom, bathroom, garage workshop, bowling alley, roadside diner, drive-in theater, construction site, amusement park and canyon town streets at golden hour. Each space has four compositions, its own floor and its own lighting. A final wide scene brings the whole town together. The set contains 45 square images at 1024 x 1024.I kept these as backgrounds only, with open floor areas and no people or living creatures. Animal-like shapes appear only as clearly inert decorative objects or clipped topiary plants. Signs and screens are blank, and I used no franchise references or readable text.Trigger: st0n34g3 sub0rbI suggest placing the exact trigger near the start of a prompt, then naming the location, viewpoint, materials and lighting. Load a trained LoRA separately through your interface; the trigger by itself does not load a model file. These are positive scene prompt examples:st0n34g3 sub0rb, prehistoric stone living room, wide doorway view, massive stone sofa, fur cushions, log coffee table, woven reed rug, golden canyon sunlight, dramatic firelight, cinematic photorealismst0n34g3 sub0rb, prehistoric roadside diner, diagonal view, slab counter, round stone stools, glowing rock-and-bone jukebox, black-and-white stone tile, warm shell lamps, red-rock canyon windowst0n34g3 sub0rb, prehistoric canyon town, elevated rooftop view, rounded stone houses, glowing windows, stone-paved street, cycads and ferns, smoking volcano, golden-hour hazeSuggested starting experiments, NOT verified settings: after training a version for a compatible conventional SDXL or Illustrious checkpoint, I would begin a comparison around 30 steps, CFG 5 and LoRA strength 0.7, then compare strengths 0.5, 0.8 and 1.0 with the same seed and prompt. These are proposed test values, not measured results or a universal recipe. Follow the actual checkpoint and sampler requirements, especially for accelerated models. For parameter meanings, I refer to the official SDXL documentation: <a target="_blank" rel="ugc" href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl">https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl</a>I have not trained or tested a LoRA from this dataset here. SDXL, Illustrious and Krea compatibility is unverified. For Krea, first establish the exact supported model, LoRA-loading workflow and exposed controls; I cannot claim that these numerical controls or this future LoRA will be available there. The images were produced with the integrated image-generation tool, not as test outputs from a trained st0n34g3 sub0rb LoRA.arrssenneFRANCAISAvec st0n34g3 sub0rb, je voulais imaginer la vie quotidienne de banlieue construite avec des dalles de pierre enormes, des troncs patines, des fibres tressees, de la fourrure et des coquillages. J'ai cree une petite ville prehistorique joyeuse dans un canyon de roche rouge, avec des maisons arrondies, des fenetres chaleureuses, des cycas et un volcan fumant a l'horizon. J'ai choisi une ambiance tres theatrale : mobilier imposant, soleil spectaculaire, ombres profondes et humour porte par les objets artisanaux anachroniques.J'ai construit ce dataset de LoRA de backgrounds autour de onze lieux distincts : salon, cuisine, chambre, salle de bain, garage-atelier, bowling, diner routier, cine-parc, chantier, fete foraine et rues de la ville a l'heure doree. Chaque lieu propose quatre compositions, un sol particulier et son propre eclairage. Une derniere vue rassemble la ville. Le set comprend 45 images carrees de 1024 x 1024.J'ai conserve uniquement des decors, avec des zones de sol degagees, sans personnes ni creatures vivantes. Les formes animales sont seulement des ornements inertes ou des arbustes tailles. Les enseignes et ecrans restent vierges, sans texte lisible ni reference a une franchise.Trigger : st0n34g3 sub0rbJe suggere de placer le trigger exact au debut du prompt, puis de preciser le lieu, le cadrage, les materiaux et la lumiere. Le LoRA entraine doit etre charge separement dans ton interface; le trigger seul ne charge aucun fichier de modele. Les exemples de prompts en anglais se trouvent dans la section anglaise ci-dessus.Points de depart proposes, NON verifies : apres entrainement pour un checkpoint conventionnel SDXL ou Illustrious compatible, je commencerais une comparaison autour de 30 steps, CFG 5 et poids LoRA 0.7, puis je comparerais les poids 0.5, 0.8 et 1.0 avec le meme seed et le meme prompt. Ce sont des valeurs d'essai proposees, pas des resultats mesures. Les exigences du checkpoint et du sampler restent prioritaires, surtout pour les modeles acceleres. Reference sur les parametres : documentation officielle SDXL, <a target="_blank" rel="ugc" href="https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl">https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl</a>Je n'ai encore entraine ni teste de LoRA a partir de ce dataset ici. La compatibilite SDXL, Illustrious et Krea reste non verifiee. Pour Krea, il faut d'abord identifier le modele exact, la prise en charge du chargement de LoRAs et les controles disponibles. Je ne confirme ni la disponibilite de ces reglages ni la compatibilite de ce futur LoRA. Les images ont ete produites avec l'outil de generation integre; elles ne sont pas des resultats de test d'un LoRA st0n34g3 sub0rb entraine.arrssenne

Files with SHA256: 69c79a533d84fb288a034dfdc8e5b0e23d0c4a58a8353324a9320f01bd3f0ed2