Skip to content
Images

Pixel Art SDXL LoRA

16×16 спрайты для игр

2 min readupdated

LoRA-адаптер для Stable Diffusion XL, который превращает текстовый промпт в одиночный игровой спрайт в стиле 16×16, увеличенный до 1024×1024. Базовая SDXL без адаптера рисует «пиксельные» иллюстрации с неровными квадратами и мягкими краями, а эта LoRA выдаёт один центрированный объект из плоских квадратных ячеек. Обучена на 1000 синтетических спрайтов из Pixel Art Atlas; меняются только веса LoRA (rank 32) в UNet, сама SDXL остаётся неизменной. Для ускорения авторы советуют совмещать её с LCM-LoRA: картинка генерируется за 6 шагов вместо 30. Подходит для предметов, простых существ и растений — но не для персонажей, сцен, тайлсетов и анимации.

Ключевые особенности

  • Один объект по центру на пиксельной сетке, похожей на 16×16.
  • Обучение на 1000 спрайтов; в датасете по одному объекту на предмет, чтобы выучить стиль, а не конкретные объекты.
  • Совместимость с LCM-LoRA: 6 шагов, ускорение до 6,7× относительно базовой SDXL.
  • В комплекте Gradio-приложение и шаг snap-to-grid для точных 16×16 спрайтов с прозрачным фоном.
  • Промпты на английском; формат: pxlsprite, pixel art sprite of <subject>, white background.

Что можно сделать

  • Генерировать спрайты предметов, зелий, сундуков, ключей, растений и простых существ для игровых прототипов.
  • Собрать небольшой набор ассетов: сгенерировать несколько сидов на объект, выбрать лучшие и прогнать через snap-to-grid.
  • Быстро проверять идеи для пиксель-арта, сочетая LoRA с LCM-LoRA для 6-шаговой генерации.
  • Запустить локальный Gradio-демо из репозитория и экспериментировать с промптами.

Пример

import torch
from diffusers import StableDiffusionXLPipeline, LCMScheduler
pipe = StableDiffusionXLPipeline.from_pretrained('stabilityai/stable-diffusion-xl-base-1.0', torch_dtype=torch.float16).to('cuda')
pipe.load_lora_weights('unsupervised-abdullah/pixel-art-sdxl-lora', weight_name='pytorch_lora_weights.safetensors', adapter_name='pixel')
pipe.load_lora_weights('latent-consistency/lcm-lora-sdxl', adapter_name='lcm')
pipe.set_adapters(['pixel', 'lcm'], adapter_weights=[1.0, 1.0])
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
image = pipe('pxlsprite, pixel art sprite of a red apple, white background', num_inference_steps=6, guidance_scale=1.0).images[0]
ПодробнееБыстрый вариант с LCM-LoRA и tiny VAE: 3,79 с на изображение, пик VRAM 8,1 ГБ, grid error 0,0190, CLIP 36,2. У базовой SDXL — 25,48 с, 9,5 ГБ, grid error 0,1236, CLIP 38,5.
Веса LoRA — около 186 МБ; обучение заняло примерно 2 часа на одной NVIDIA T4.
Ограничения: фон не всегда белый, слабые результаты для мечей и пончиков, сырой вывод лишь близок к сетке; нужен snap-to-grid. Модель не для персонажей, сцен, тайлсетов, анимации и спрайтов больше 16×16.
Discuss Telegram 0 6

Will it run?

Runs in F32

Fits without quantisation.

  • F32weights 9.7 GB · ready files12 GB
  • BF16weights 5.0 GB7.0 GB
  • FP8 / INT8recommendedweights 2.6 GB4.3 GB
  • 4-bitweights 1.5 GB3.2 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · F32 — from RTX 5080

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

This is a LoRA: counted together with its base model stable-diffusion-xl-base-1.0.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # unsupervised-abdullah/pixel-art-sdxl-lora
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download stabilityai/stable-diffusion-xl-base-1.0 --local-dir ~/models/stable-diffusion-xl-base-1.0
    hf download unsupervised-abdullah/pixel-art-sdxl-lora --local-dir ~/models/pixel-art-sdxl-lora
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • SageAttention 3v2.2.0from October 28, 2025

    even faster on RTX 50 / Blackwell

    FP4 attention for Blackwell. Faster than SageAttention 2, but the authors suggest v2 when accuracy matters.

    How to enable: Build from the sageattention3_blackwell folder of the repo; in ComfyUI via KJNodes.

    Earlier generations (2)
    • SageAttention 2++ — Quantised (INT8/FP8) attention with no visible quality loss. The main speed-up for image and video generation.
    • SageAttention 1 — First version: INT8 for Q·K. Still works, but 2++ is faster and more accurate.
  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (3)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
    • xFormers — The older way to speed up attention, works almost everywhere. Sage or Flash are faster where available.
  • cache-dit (DBCache)v1.5.6from November 20, 2026new release

    1.5–3× at nearly the same quality

    Reuses block outputs between diffusion steps. Supports most DiT models.

    How to enable: pip install cache-dit; one line cache_dit.enable_cache(pipe) for diffusers.

    Earlier generations (2)
    • MagCache — Skips steps by magnitude of change. More accurate than TeaCache, no calibration.
    • TeaCache — The first popular diffusion step cache. Works, newer methods are more accurate.
  • Nunchaku (SVDQuant 4-bit)v1.2.1from January 25, 2026

    ≈3× faster and 3.5× less memory

    4-bit versions of popular image models with almost no quality loss (FP4 on RTX 50).

    How to enable: Ready weights for supported models; in ComfyUI use ComfyUI-nunchaku.

  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

Requires the base model

stabilityai/stable-diffusion-xl-base-1.0

This model is a LoRA / adapter. Download the base model first, then load this one on top.

New posts about versions of stable-diffusion-xl-base-1.0 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 5All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face
Merges 2All on Hugging Face

About the model

unsupervised-abdullah/pixel-art-sdxl-lora
Author
unsupervised-abdullah
Category
Images
Released
Posted

On NeyroLearn

Hugging Face

60

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments