Views
No views yet
runwayml/stable-diffusion-v1-5 for photorealistic
interior renovation image generation.| Parameter | Value |
|---|---|
| Base Model | runwayml/stable-diffusion-v1-5 |
| LoRA Rank | 16 |
| Epochs | 4 |
| Training Images | 3000 |
| Avg Loss Epoch 1 | 0.1548 |
| Avg Loss Epoch 2 | 0.1510 |
| Avg Loss Epoch 3 | 0.1379 |
| Avg Loss Epoch 4 | 0.1424 |
| Best Loss | 0.0022 |
| Metric | Value |
|---|---|
| Avg CLIP Score | 0.2919 |
| Style Classification Accuracy | 94.17% |
1from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler
2from peft import PeftModel
3import torch
4
5pipe = StableDiffusionPipeline.from_pretrained(
6 'runwayml/stable-diffusion-v1-5',
7 torch_dtype=torch.float16
8).to('cuda')
9
10pipe.unet = PeftModel.from_pretrained(pipe.unet, 'Mrumairahmad100/renovision-interior-lora')
11pipe.unet = pipe.unet.merge_and_unload()
12pipe.scheduler = DPMSolverMultistepScheduler.from_config(
13 pipe.scheduler.config,
14 algorithm_type='dpmsolver++', solver_order=2
15)
16
17img = pipe(
18 prompt='a beautifully renovated modern living room interior, photorealistic, 8K quality, no people',
19 negative_prompt='ugly, blurry, outdoor, people, watermark',
20 num_inference_steps=30,
21 guidance_scale=7.5
22).images[0]