ImagePulseV2 数据集-多分辨率

# ImagePulseV2 数据集-多分辨率

ImagePulseV2 数据集是我们为训练 Diffusion Templates 系列模型而构造的数据集，数据集包含多个子集，根据 DiffusionDB 中随机采样的提示词，使用 Z-Image-Turbo、Qwen-Image、Qwen-Image-Edit 等模型生成得到。

* 开源代码：[DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio)
* 技术报告：[arXiv](https://arxiv.org/abs/2604.24351)
* 项目主页：[GitHub](https://modelscope.github.io/diffusion-templates-web/)
* 文档参考：[English Version](https://diffsynth-studio-doc.readthedocs.io/en/latest/Diffusion_Templates/Introducing_Diffusion_Templates.html)、[中文版](https://diffsynth-studio-doc.readthedocs.io/zh-cn/latest/Diffusion_Templates/Introducing_Diffusion_Templates.html)
* 在线体验：[魔搭社区创空间](https://modelscope.cn/studios/DiffSynth-Studio/Diffusion-Templates)
* 模型集：[ModelScope](https://modelscope.cn/collections/DiffSynth-Studio/KleinBase4B-Templates)、[ModelScope 国际站](https://modelscope.ai/collections/DiffSynth-Studio/KleinBase4B-Templates)、[HuggingFace](https://huggingface.co/collections/DiffSynth-Studio/kleinbase4b-templates)
* 数据集：[ModelScope](https://modelscope.cn/collections/DiffSynth-Studio/ImagePulseV2)、[ModelScope 国际站](https://modelscope.cn/collections/DiffSynth-Studio/ImagePulseV2)、[HuggingFace](https://huggingface.co/collections/DiffSynth-Studio/imagepulsev2)

## 数据样例

|描述|宽高|图像|
|-|-|-|
|花园里，一只毛茸茸的小兔子正蹲坐在柔软的草地上，阳光透过树叶洒下斑驳光影。它头上戴着一顶造型滑稽的彩色小帽子，帽檐歪斜，仿佛刚从树梢上跳下来。背景是繁茂的绿植与几朵盛开的野花，远处隐约可见木栅栏和几片飘动的蒲公英。画面采用50mm镜头拍摄，景深柔和，前景清晰，背景略带虚化，整体色调温暖明亮，充满童话般的童趣与宁静。|768 $\times$ 1024||
|The image is a stylized, digitally rendered portrait featuring a young woman with a cheerful, expressive face, set against a softly lit, dreamlike background. She has large, bright eyes, a gentle smile, and her hair is styled in soft waves, often accented with pastel or glittery tones. She is dressed in a fashionable, contemporary outfit—possibly a sequined top or a light, flowing dress—with subtle details like ruffles or lace. The lighting is warm and diffused, casting a glow around her, enhancing the ethereal, almost cinematic quality of the scene.\n\nThe background is a blend of abstract and surreal elements: floating orbs of light, soft gradients of pastel colors, and faint, impressionistic brushstrokes reminiscent of Thomas Kinkade’s signature style. There are no visible architectural structures or realistic environments—instead, the scene feels like a curated, fantasy-inspired dreamscape, evoking the whimsical and romantic aesthetics often associated with Beeple’s digital art.\n\nThe overall composition is centered on the subject, with a shallow depth of field that blurs the background slightly to emphasize her presence. The color palette is predominantly soft pinks, blues, and golds, contributing to a nostalgic, fairy-tale atmosphere. No other characters or objects are present—this is a solo portrait, focused entirely on the emotional and visual impact of the subject.|1920 $\times$ 1088||
|A highly detailed, ultra-detailed anime-style digital illustration of Clementine from The Walking Dead, rendered with intricate linework and dynamic lighting. She wears her iconic baseball cap featuring the letter “D” emblazoned on the front, slightly tilted to one side, casting a soft shadow over her eyes. Her expression is determined yet calm, with sharp, defined facial features and expressive eyes that convey resilience. Her hair is neatly tied back, with strands softly framing her face under the cap’s brim. The lighting is dramatic, with highlights catching the edges of her cap, the contours of her cheekbones, and the fabric of her clothing, creating a cinematic depth. The background is minimal yet atmospheric, with subtle gradients and faint environmental details that suggest an outdoor setting, possibly dusk or dawn, enhancing the emotional tone. The art style blends the precision of Artgerm’s character design with the painterly textures and ethereal lighting of Greg Rutkowski, and the elegant, ornamental composition reminiscent of Alphonse Mucha. Every strand of hair, stitch on her clothing, and fold in the fabric is rendered with meticulous sharp focus, capturing the essence of a digital painting created on an art station. The composition is balanced and symmetrical, emphasizing Clementine’s poised stance and the symbolic weight of her iconic hat.|864 $\times$ 1152||

# ImagePulseV2 数据集-多分辨率

ImagePulseV2 数据集是本团队为训练**扩散模板（Diffusion Templates）**系列模型所构建的数据集，该数据集包含多个子集，其基于从DiffusionDB中随机采样的提示词，借助Z-Image-Turbo、Qwen-Image、Qwen-Image-Edit等模型生成所得。

* 开源代码：[DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio)
* 技术报告：[arXiv](https://arxiv.org/abs/2604.24351)
* 项目主页：[GitHub](https://modelscope.github.io/diffusion-templates-web/)
* 文档参考：[英文版本](https://diffsynth-studio-doc.readthedocs.io/en/latest/Diffusion_Templates/Introducing_Diffusion_Templates.html)、[中文版本](https://diffsynth-studio-doc.readthedocs.io/zh-cn/latest/Diffusion_Templates/Introducing_Diffusion_Templates.html)
* 在线体验：[魔搭社区创空间](https://modelscope.cn/studios/DiffSynth-Studio/Diffusion-Templates)
* 模型集：[ModelScope](https://modelscope.cn/collections/DiffSynth-Studio/KleinBase4B-Templates)、[ModelScope国际站](https://modelscope.ai/collections/DiffSynth-Studio/KleinBase4B-Templates)、[HuggingFace](https://huggingface.co/collections/DiffSynth-Studio/kleinbase4b-templates)
* 数据集：[ModelScope](https://modelscope.cn/collections/DiffSynth-Studio/ImagePulseV2)、[ModelScope国际站](https://modelscope.cn/collections/DiffSynth-Studio/ImagePulseV2)、[HuggingFace](https://huggingface.co/collections/DiffSynth-Studio/imagepulsev2)

## 数据样例

|描述|宽×高|图像|
|-|-|-|
|花园里，一只毛茸茸的小兔子正蹲坐在柔软的草地上，阳光透过树叶洒下斑驳光影。它头上戴着一顶造型滑稽的彩色小帽子，帽檐歪斜，仿佛刚从树梢上跳下来。背景是繁茂的绿植与几朵盛开的野花，远处隐约可见木栅栏和几片飘动的蒲公英。画面采用50mm镜头拍摄，景深柔和，前景清晰，背景略带虚化，整体色调温暖明亮，充满童话般的童趣与宁静。|768 $ imes$ 1024||
|本图像为风格化数字渲染肖像，主体是一位面容开朗灵动的年轻女性，背景光线柔和如梦似幻。她拥有明亮的大眼睛与温柔的笑容，发丝呈柔和波浪造型，常搭配马卡龙色调或闪粉装饰。身着时尚现代的服饰——可能是亮片上衣或轻盈飘逸的连衣裙，带有荷叶边、蕾丝等精致细节。整体光线温暖柔和，在她周身晕开柔光，强化了画面空灵且近乎电影质感的氛围。

背景融合抽象与超现实元素：浮动的光球、柔和的马卡龙色渐变，以及隐约带有托马斯·金凯德（Thomas Kinkade）标志性风格的印象派笔触。画面无可见的建筑结构或写实场景，整体更像是精心打造的奇幻梦境，唤起与比普尔（Beeple）数字艺术常关联的奇趣浪漫美学风格。

整体构图以主体为中心，采用浅景深效果轻微虚化背景以突出人物存在感。色调以柔和的粉、蓝、金为主，营造出怀旧童话般的氛围。画面无其他人物或物体，为单人肖像，完全聚焦于主体的情感与视觉表现力。|1920 $ imes$ 1088||
|本作品为《行尸走肉》中克莱曼婷（Clementine）的超精细动漫风格数字插画，线条繁复精致，光影富有动感。她头戴标志性棒球帽，帽正面印有字母“D”，略微偏向一侧，在眼部投下柔和阴影。她神情坚毅却平和，面部轮廓清晰分明，眼神灵动传递出坚韧的气质。头发整齐向后梳，发丝在帽檐下轻柔地勾勒出脸部轮廓。光线极具戏剧感，高光落在帽檐、颧骨轮廓与衣物面料上，营造出电影级的景深效果。背景极简却富有氛围感，带有柔和渐变与隐约的环境细节，暗示户外场景，可能为黄昏或黎明，强化了情绪基调。艺术风格融合了阿特杰姆（Artgerm）角色设计的精准性、格雷格·鲁特科夫斯基（Greg Rutkowski）的绘画质感与空灵光影，以及阿尔丰斯·穆夏（Alphonse Mucha）标志性的优雅装饰性构图。每一缕发丝、衣物上的针脚与布料褶皱都以极致精细的焦点渲染，完美呈现了数位绘画创作的精髓。构图均衡对称，突出克莱曼婷沉稳的姿态与标志性帽子的象征意义。|864 $ imes$ 1152||

搜集汇总

背景与挑战

背景概述

ImagePulseV2-多分辨率数据集专为训练Diffusion Templates系列模型而构建，通过从DiffusionDB随机采样提示词，并利用Z-Image-Turbo、Qwen-Image等模型生成图像，包含多个子集以支持多分辨率应用。