CoolFace
Modelpublic

manupm87/finetuned-sd-1.4-oldbookillustrations-lora

sourceHugging Facecreativeml-openrail-mupdated 3mo agoView on Hugging Face
0likes18downloads
Model Card

LoRA - Estilo de ilustraciones de libros antiguos (SD v1-4)

LoRA entrenada sobre `CompVis/stable-diffusion-v1-4` con el dataset `gigant/oldbookillustrations` para reproducir el estilo de las ilustraciones de libros antiguos.

🔁 Versión de finetuning completo (pipeline entero con la UNet finetuneada): `manupm87/finetuned-sd-1.4-oldbookillustrations`.

Trigger de estilo: anade old book illustration, engraving, antique al final del prompt para activar el estilo.

Uso

python
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("manupm87/finetuned-sd-1.4-oldbookillustrations-lora")

prompt = "a lighthouse on a cliff, old book illustration, engraving, antique"
image = pipe(prompt).images[0]
image.save("output.png")

Detalles del entrenamiento

ParametroValor
Modelo baseCompVis/stable-diffusion-v1-4
Datasetgigant/oldbookillustrations (columnas 1600px / info_alt)
MetodoLoRA en la UNet (capas de atencion)
Rango / alpha16 / 16
Epocas3
Learning rate0.0001
Batch size6
Resolucion512x512
Trigger de estiloold book illustration, engraving, antique

Ejemplos

Generadas con esta LoRA (con el trigger de estilo, misma semilla):

PromptImagen
A sailing ship with billowing sails rides the tall waves of a stormy sea<img src="samples/sample_0.png" width="256">
A ruined castle stands on a cliff overlooking the misty sea at dusk<img src="samples/sample_1.png" width="256">
A young girl wanders alone into a forest of tall, twisted trees<img src="samples/sample_2.png" width="256">
A solitary lighthouse rises above the rocky shore as waves break below it<img src="samples/sample_3.png" width="256">
A steam locomotive races across the countryside, trailing a long plume of smoke<img src="samples/sample_4.png" width="256">