Diffusers
stable-diffusion
lora
style

Style LoRA: Simpsons cartoon style without a trigger word (SD 1.4)

NYCU Image and Video Generation 2026 — Lab2 Task 2-1 (student 113550175). Non-commercial coursework.

  • Base model: CompVis/stable-diffusion-v1-4
  • Data: all 755 images of Norod78/simpsons-blip-captions (CC-BY-NC-SA-4.0). The BLIP captions were rewritten to remove every "simpson"/"The Simpsons" mention, so the style is applied to ordinary prompts instead of being tied to the show's name.
  • Training: train_lora.py (via train_lora_custom.sh), 512 px, batch 4, 20 epochs (3780 steps), lr 1e-4 constant, fp16, random flip, seed 42, UNet LoRA
import torch
from diffusers import StableDiffusionPipeline
pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", torch_dtype=torch.float16).to("cuda")
pipe.load_lora_weights("HuaNG038B/sd14-lora-simpsons-style")
pipe("a man with sunglasses", num_inference_steps=30, guidance_scale=7.5).images[0]
Downloads last month
34
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for HuaNG038B/sd14-lora-simpsons-style

Adapter
(640)
this model

Dataset used to train HuaNG038B/sd14-lora-simpsons-style