A LoRA fine-tune of Stable Diffusion XL 1.0 that generates images in the style of classic ANSI art from the BBS era (1990s). Trained on rendered ANSI/RIP art from the ACiD Productions art packs.
1from diffusers import StableDiffusionXLPipeline
2import torch
3
4pipe = StableDiffusionXLPipeline.from_pretrained(
5 "stabilityai/stable-diffusion-xl-base-1.0",
6 torch_dtype=torch.float16,
7).to("cuda")
8
9pipe.load_lora_weights("cahlen/acid-ansi-lora")
10
11prompt = "acid-ansi-style, a menacing skull wreathed in flames on a black background"
12negative_prompt = "blurry, photorealistic, photo, smooth gradients, 3d render, watermark"
13
14image = pipe(
15 prompt,
16 negative_prompt=negative_prompt,
17 num_inference_steps=30,
18 guidance_scale=7.5,
19 cross_attention_kwargs={"scale": 0.7},
20).images[0]
21
22image.save("output.png")
The LoRA scale controls style strength at inference time. Lower values (0.5-0.6) preserve more of the base model's composition while higher values (0.8-1.0) push harder into the ANSI style but may cause repetition artifacts. 0.7 is the recommended default.
All samples generated at step 3500, LoRA scale 0.7, seed 42.
276 images derived from 50 curated ANSI/RIP art renders from ACiD Productions art packs (1993-1996):
