Downloads · 30 days
43
39% of all-time downloads
wwwwwwwwwwwwwwwwwfdd/lai-sd15-lora-v1
lai-sd15-lora-v1 is a text-to-image model from wwwwwwwwwwwwwwwwwfdd. Use it when you need an image from a text prompt. It is set up for diffusers. The card lists the license as creativeml-openrail-m.
Multi-domain SD 1.5 LoRA fine-tuned on 780K samples across 16 datasets (faces, animals, art, anime/characters, food, textures, Danbooru). Rank 128, trained for 600 steps at 512² on a single NVIDIA B200.
Downloads · 30 days
43
39% of all-time downloads
All-time downloads
110
Public
Repo size
111 MB
Likes
1
Public
Click a slice to open those files.
.safetensors102 MB · 92%
From the Hugging Face model README
Multi-domain SD 1.5 LoRA fine-tuned on 780K samples across 16 datasets (faces, animals, art, anime/characters, food, textures, Danbooru). Rank 128, trained for 600 steps at 512² on a single NVIDIA B200.
| metric | base SD 1.5 | LAI (ours) |
|---|---|---|
| mean CLIP ↑ | 0.301 | 0.304 |
| Inception Score ↑ | 1.67 ± 0.16 | 1.73 ± 0.16 |
| FID vs training ↓ | 375.1 | 364.0 |
| category | base | LAI | Δ |
|---|---|---|---|
| faces | 0.287 | 0.283 | -0.004 |
| animals | 0.337 | 0.331 | -0.006 |
| art | 0.332 | 0.327 | -0.005 |
| characters | 0.293 | 0.310 | +0.017 |
| food | 0.256 | 0.267 | +0.011 |
Biggest per-prompt win: a pokemon creature, colorful background → +0.030 CLIP.
from diffusers import StableDiffusionPipeline
import torch
pipe = StableDiffusionPipeline.from_pretrained(
"stable-diffusion-v1-5/stable-diffusion-v1-5",
torch_dtype=torch.float16,
).to("cuda")
pipe.load_lora_weights("wwwwwwwwwwwwwwwwwfdd/lai-sd15-lora-v1")
img = pipe(
"a pokemon creature, colorful background, detailed",
negative_prompt="low quality, blurry, bad anatomy, extra fingers",
num_inference_steps=25,
guidance_scale=7.0,
).images[0]
img.save("out.png")
ComfyUI: drop into models/loras/, strength 0.7-1.0.
AUTOMATIC1111: drop into models/Lora/, reference as <lora:lai-sd15-lora-v1:0.8>.
stable-diffusion-v1-5/stable-diffusion-v1-5CreativeML Open RAIL-M (inherits from SD 1.5).