Pixel Art XL

Pixel Art XL

A pixel art style add-on for Stable Diffusion XL that turns text prompts into pixel-style images, with an optional LCM LoRA for faster runs. For game makers, designers and hobbyists.

Made by nerijs on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Pixel Art XL is a LoRA for the Stable Diffusion XL base model that makes text-to-image results look like pixel art. It needs no style prompt and no trigger keyword, and the card says it works with both the 0.9 and 1.0 versions of the base model. You can pair it with an LCM LoRA to generate faster. The card lists v1 as its first release.

Who it suits: It suits people who want pixel art from a text prompt without drawing it by hand.

Industry | Who uses it | How they use it | Example
Games | Indie game makers | Draw sprites and scenes in a pixel style | Prompt for a character and downscale the result
Design | Graphic designers | Try retro looks for posters, icons and social posts | A set of nine pixel images from one prompt
Hobby | Digital artists | Explore pixel art ideas quickly | Prompt a small animal and compare variations
Software | Developers | Add pixel image generation to a tool | Call the pipeline from a Python script

If you want realistic or 3D renders, you can skip it, since it is built for pixel art.

Where to use it: You run it in Python with the diffusers library on top of Stable Diffusion XL. The card's example moves the pipeline to a CUDA graphics card.

• Python scripts that use diffusers
• A machine with an NVIDIA GPU and CUDA
• Windows, macOS or Linux with the huggingface_hub package

Use case | Who | What they get back
Pixel art images | Artists and game makers | Pixel-style pictures from a text prompt
Fast batches | Developers | Several images quickly with the LCM LoRA pairing
Isometric scenes | Designers | The card says it works with isometric and non-isometric styles

How to set it up: You need Python, the diffusers and torch libraries, the Stable Diffusion XL base model, and the Pixel Art XL weights file saved in your working folder. The card's example runs on a CUDA graphics card.

• Python on Windows, macOS or Linux
• The diffusers and torch libraries
• An NVIDIA GPU with CUDA for the card's example

1. Install the Hugging Face hub package.
2. Download the Pixel Art XL files.
3. Put the weights file where the script expects it, then run the script.

pip install -U huggingface_hub
hf download nerijs/pixel-art-xl

How to use it: Write a prompt that starts with the word pixel, add a negative prompt, and generate. The card suggests 8 steps, a guidance scale of 1.5 and a strength of 1.2 for this LoRA when you pair it with an LCM LoRA. Downscale the result 8 times with nearest neighbors for pixel-perfect images.

from diffusers import DiffusionPipeline, LCMScheduler
import torch

model_id = "stabilityai/stable-diffusion-xl-base-1.0"
lcm_lora_id = "latent-consistency/lcm-lora-sdxl"
pipe = DiffusionPipeline.from_pretrained(model_id, variant="fp16")
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)

pipe.load_lora_weights(lcm_lora_id, adapter_name="lora")
pipe.load_lora_weights("./pixel-art-xl.safetensors", adapter_name="pixel")

pipe.set_adapters(["lora", "pixel"], adapter_weights=[1.0, 1.2])
pipe.to(device="cuda", dtype=torch.float16)

prompt = "pixel, a cute corgi"
negative_prompt = "3d render, realistic"

num_images = 9

for i in range(num_images):
    img = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        num_inference_steps=8,
        guidance_scale=1.5,
    ).images[0]

    img.save(f"lcm_lora_{i}.png")

Tool | What it does
LCMScheduler | Lets the pipeline run in few steps with the LCM LoRA
load_lora_weights | Loads the LCM LoRA and the Pixel Art XL weights
set_adapters | Sets the strength of each adapter, 1.0 and 1.2 in the card
nearest neighbors downscale | Shrinks the output 8 times for pixel-perfect images

Example: You prompt "pixel, a cute corgi" with a negative prompt of "3d render, realistic". The script saves nine images, each made in 8 steps, as numbered image files in the working folder.

For newcomers and experts: If you are new to this, a LoRA is a small add-on that changes the look of a bigger image model. A first small step is to run the card's script and open the saved images. Use a fixed VAE if you see artifacts.

If you know the field, this LoRA sits on SDXL base, and the card advises against the refiner. It also says the model works well with only one text encoder. The LCM pairing trades some control for speed, at 8 steps and a guidance scale of 1.5.

Text prompt with the word pixel
SDXL base model plus two LoRAs
Eight steps with the LCM scheduler
Downscale 8 times, nearest neighbors
A pixel-perfect image

Good to know: The card suggests a fixed VAE, either the 0.9 version or the fp16 fix, to avoid artifacts. The weights file in the script is loaded from the current folder.

• License: creativeml-openrail-m
• Cost to run: no API key or paid service is named, but the example needs a CUDA graphics card
• Data it reaches: the download comes from Hugging Face, and the model works on the prompts you give it on your own machine
• Not for: use with the SDXL refiner, which the card says not to use

On Hugging Face, 14 November 2026

• Task: text to image
• Library: diffusers
• Downloads in the last 30 days: 38,781
• Likes: 669
• License: creativeml-openrail-m
• Based on: stabilityai/stable-diffusion-xl-base-1.0
• Tags: diffusers, text-to-image, stable-diffusion, lora

Links

• Model: https://huggingface.co/nerijs/pixel-art-xl
• Files: https://huggingface.co/nerijs/pixel-art-xl/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download nerijs/pixel-art-xl

Then follow the usage in its model card. Nothing in this kit runs by itself.