ac-blackflag-style-lora

AC Black Flag Style LoRA

A LoRA fine-tune of Stable Diffusion 1.5's UNet that repaints a photo in the art style of Assassin's Creed Black Flag. Use it with img2img: upload a photo, get back a version of yourself painted like a pirate out of the game.

Trained by Bakeer as part of the weekend_ai project, alongside captain-byte-qwen2.5-7b and the companion ac-blackflag-character-classifier.

Usage

import torch
from diffusers import StableDiffusionImg2ImgPipeline
from PIL import Image

pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5", torch_dtype=torch.float16
)
pipe.load_lora_weights("SkullSameh/ac-blackflag-style-lora")
pipe.to("cuda")

image = Image.open("your_photo.jpg").convert("RGB").resize((512, 512))
result = pipe(
    prompt="concept art in the style of Assassin's Creed Black Flag, painterly pirate video game character portrait",
    image=image,
    strength=0.6,
    guidance_scale=7.5,
).images[0]
result.save("stylized.png")

strength controls how much of your original photo is preserved: lower values (~0.4-0.5) keep more of your likeness and background, higher values (~0.7-0.9) lean further into the art style at the cost of resemblance.

Training details

• Base model: stable-diffusion-v1-5/stable-diffusion-v1-5
• Method: LoRA (r=16, alpha=32) on the UNet's attention projections (to_q, to_k, to_v, to_out.0), trained via PEFT + diffusers
• Data: a small set of Assassin's Creed Black Flag character reference images, all captioned with a single shared style prompt
• This is an early, small-dataset version -- more reference images will improve style consistency in future updates.