🏠 Stable Diffusion v1.5 — LoRA Fine-tune (Interior Design, Living Rooms)
- Prompt
- Modern living room with a red sofa and soft lighting in a vintage style.
- Negative Prompt
- blurry, distorted, low quality
- Prompt
- Spacious living room with a leather sofa and a glass coffee table.
- Negative Prompt
- blurry, distorted, low quality
- Prompt
- Cozy living room with a fireplace, wooden floors, and neutral tones.
- Negative Prompt
- blurry, distorted, low quality
- Prompt
- Luxurious living room with gold accents and a large chandelier.
- Negative Prompt
- blurry, distorted, low quality
- Prompt
- Minimalist living room with white walls and a single gray sofa.
- Negative Prompt
- blurry, distorted, low quality
- Prompt
- Rustic living room with exposed wooden beams and stone wall accents.
- Negative Prompt
- blurry, distorted, low quality
- Prompt
- Living room with a large sectional sofa and soft blue tones.
- Negative Prompt
- blurry, distorted, low quality
- Prompt
- Contemporary living room with bold geometric patterns and vibrant colors.
- Negative Prompt
- blurry, distorted, low quality
- Prompt
- Living room with an eclectic mix of furniture and a gallery wall.
- Negative Prompt
- blurry, distorted, low quality
- Prompt
- Industrial living room with exposed brick walls and metal furniture.
- Negative Prompt
- blurry, distorted, low quality
This model is a LoRA fine-tune of Stable Diffusion v1.5 (sd-v1-5-pruned-noema-fp16.safetensors) trained to generate photorealistic and style-consistent interior design images, with a focus on living rooms.
📊 Training Setup
- Method: DreamBooth + LoRA (via sd-scripts)
- Resolution: 512×512
- Batch size: 2
- Epochs: ~8 per style
- LoRA config:
network_dim = 16,network_alpha = 8 - Optimizer: AdamW8bit
- Learning rates:
unet = 5e-4,text_encoder = 1e-4 - Scheduler: cosine with restarts
- Hardware: NVIDIA T4 (Google Colab Pro)
- Precision: mixed FP16
🗂️ Dataset
Custom dataset curated from 3 public sources, filtered to living room interiors.
Covers 7 interior design styles:
- Coastal
- Industrial
- Mid-Century Modern
- Modern
- Rustic
- Scandinavian
- Traditional
Each style was fine-tuned separately, then merged into one balanced checkpoint (merged_lora.safetensors).
📐 Evaluation
Evaluation was performed on 350 generated images with CLIP score (prompt–image alignment) and FID (Fréchet Inception Distance) for visual realism.
- Overall FID: 88.93 (lower is better, <50 = photorealistic)
- Style-specific FID:
- Coastal — 160.45 (best)
- Traditional — 176.21
- Rustic — 177.91
- Scandinavian — 177.96
- Modern — 181.65
- Mid-Century Modern — 182.27
- Industrial — 188.65
ℹ️ While FID > 50 means results are not “perfectly photorealistic,” qualitative inspection shows this model produces stable and style-coherent room renderings across all seven design styles.
🎯 Trigger Words
Recommended keywords:
interior, living room, house design
Combine with a style for best results:
- “modern living room interior, cozy house design, ultra realistic”
- “rustic interior living room, warm lighting, photorealistic”
🚀 Usage (🤗 Diffusers)
from diffusers import StableDiffusionPipeline
import torch
base = "runwayml/stable-diffusion-v1-5"
pipe = StableDiffusionPipeline.from_pretrained(base, torch_dtype=torch.float16).to("cuda")
# load this LoRA
pipe.load_lora_weights("aimhkim174/sd15-lora-livingroom")
image = pipe("modern living room interior design, Scandinavian style, ultra realistic",
num_inference_steps=30, guidance_scale=7.5).images[0]
image.save("out.png")
- Downloads last month
- 3
Model tree for aimhkimi74/sd15-lora-LivingRoom
Base model
runwayml/stable-diffusion-v1-5