nerijs/pixelparti-128-v0.1

---
license: apache-2.0
dataset_info:
features:
- name: image
dtype: image
splits:
- name: train
num_bytes: 34521297.512
num_examples: 4836
download_size: 34764278
dataset_size: 34521297.512
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---
<img src="https://cdn-uploads.huggingface.co/production/uploads/6303f37c3926de1f7ec42d3e/Ufp7Ec0nfJblft2VXXZBa.png" width="512" height="512">

**PixelParti128** is a synthetic dataset derived from SDXL Turbo and Pixel Art XL.
Contains 4 samples of each prompt in <a href="https://github.com/google-research/parti">PartiPrompts</a>

PixelParti128 v0.1 is a collection of ~4800 synthetic samples with their corresponding prompt and seed. Each sample is downscaled and quantized to 128px making them pixel perfect.

## Replication
You can replicate this dataset using SDXL Turbo with Pixel Art XL, 8 steps of Euler a, CFG = 1 and Diffusers.

<img src="https://cdn-uploads.huggingface.co/production/uploads/6303f37c3926de1f7ec42d3e/GUDdifE0G2NQxOTX9zl0t.jpeg" width="512" height="512">
<img src="https://cdn-uploads.huggingface.co/production/uploads/6303f37c3926de1f7ec42d3e/rgEcDVHYjRPLf1NZ5Hssq.jpeg" width="512" height="512">
<img src="https://cdn-uploads.huggingface.co/production/uploads/6303f37c3926de1f7ec42d3e/yJEWENiVtNqO-yr7Kp4ul.jpeg" width="512" height="512">

原始信息汇总

数据集概述

基本信息

• 名称: PixelParti128 v0.1
• 类型: 合成数据集
• 来源: 基于SDXL Turbo和Pixel Art XL生成
• 样本数量: 约4800个样本
• 特征: 每个样本包含对应的提示和种子信息
• 图像处理: 所有样本均被缩放和量化至128像素，确保像素完美

数据集结构

• 特征:
  • 名称: image
  • 数据类型: image
• 分割:
  • 训练集:
    • 样本数量: 4836
    • 数据大小: 34521297.512字节
    • 下载大小: 34764278字节

许可

• 许可证: Apache-2.0

配置

• 配置名称: default
• 数据文件:
  • 分割: train
  • 路径: data/train-*

数据集复制方法

• 使用SDXL Turbo结合Pixel Art XL，采用8步Euler a方法，配置CFG = 1，通过Diffusers工具实现数据集的复制。