LEOSAM's HelloWorld XL

2024.1.31 HelloWorld 4.0 version updateHelloWorld4.0 is a progressive transitional version from tagging with blip+clip to tagging with GPT4V. I initially trained a pure GPT4V tagging model, and then merged it with a large proportion of the HelloWorld3.2 version and a small proportion of other models. The new version has shown improvements in prompt compliance and concept coverage compared to the 3.2 version.The new GPT4V tagging training set has doubled from the 4000 images of the helloworld3 series to 8000 images, covering not only portraits but also animals, architecture, nature, food, illustrations, and more. However, the pure GPT4V version encountered an overfitting problem, which is preliminarily attributed to the doubling of the number of training images. One of the next steps in iterative optimization is to find out how to include as many non-portrait concepts as possible while ensuring sufficient training of portraits. At this stage, a fusion of the new and old versions has been used for fine-tuning to ensure a smooth transition between versions, so the expanded concept set and the advantages brought by GPT4V tagging are not very perceptible at the moment. These advantages will become increasingly apparent in the subsequent generations 5 and 6 of the model.2024.1.31 HelloWorld 4.0版本更新HelloWorld4.0是从blip＋clip打标到GPT4V打标的渐进过渡版本，采用最新GPT4V打标训练出的HelloWorld模型与原HelloWorld3.2进行了融合调优。新版本在提示词跟随性与概念覆盖度上相比3.2版本有所提升。新的GPT4V打标训练集在helloworld3系列4000张基础上翻了一倍，达到了8000张，内容除人像外，还涵盖动物、建筑、自然、美食、插画等内容。但纯GPT4V版本遇到了过拟合的问题，初步判断是训练图数量翻倍导致。如何在保证人像充分训练前提下，尽可能加入更多人像外其他概念，是下一步的迭代优化方向之一。现阶段就先使用了新旧版本融合调优来保证版本间平稳过渡，因此概念集扩充以及GPT4V打标带来的优势目前感知不会非常强。这些优势将在后续5代及6代模型中愈加明显体现出来。

leosamsHelloworldXL_helloworldXL40.safetensors

HelloWorld XL 4.0

HelloWorld XL 7.0

HelloWorld XL 6.0

HW5.0_Euler_a_Lightning

HelloWorld XL 5.0 GPT4V

HelloWorld XL 3.2 DPO

HelloWorld XL 3.0

HelloWorld XL 2.0

HelloWorld XL 1.0

Film Grain 2.0

Film Grain 1.0

Utopia 3.0

Utopia 2.0

Utopia

Reality 3.0

Fantasy2.0

Reality2.0

Reality

Fantasy

LEOSAM HelloWorld | SDXL Realism Checkpoint

2024.1.22 HelloWorld 4.0 Version UpdateHelloWorld 4.0 is an incremental transition version from blip+clip to GPT4V tagging. It integrates the latest HelloWorld model trained with GPT4V tagging with the original HelloWorld 3.2, and in the final round of integration, incorporates a 0.05 ratio of Juggernaut XL to adjust skin tone. The new version shows improvements in prompt word coherence and conceptual coverage compared to version 3.2.The new GPT4V tagging training set doubles the size of the HelloWorld3 series from 4000 to 8000 images, covering not only portraits but also animals, buildings, nature, food, illustrations, and more. However, the pure GPT4V version encountered overfitting issues, likely due to doubling the training image quantity. The next iterative optimization direction will focus on adding more non-portrait concepts while ensuring sufficient portrait training. Currently, a hybrid tuning of the new and old versions is used to ensure a smooth transition between versions, hence the expanded concept set and advantages brought by GPT4V tagging may not be very prominent. These advantages will become more apparent in the subsequent 5th and 6th generation models.Special Thanks:Special thanks to the development team of the <a href="https://github.com/jiayev/GPT4V-Image-Captioner" target="_blank">GPT4V Image Tagging and Processing Toolbox</a> (Jiaye, SleeepyZhou, Fok, 十字鱼). GPT4V tagging will be one of the technical foundations for future upgrades of this model. We have also developed a <a href="https://github.com/SleeeepyZhou/sd-webui-GPT4V-Image-Captioner" target="_blank">web-based plugin version</a> for everyone to explore and use.Thanks to Yuzi for developing the <a href="https://greasyfork.org/zh-CN/scripts/484783-unsplash%E5%9B%BE%E9%9B%86%E4%B8%8B%E8%BD%BD%E5%99%A8" target="_blank">Unsplash Collection Downloader Greasemonkey Script</a>, Unsplash is the world's largest royalty-free image sharing community.Thanks to Maoruoyu, SleeepyZhou, Fok, and Cayden for sharing my collection of images. The Human Completion Plan has a long way to go, and I will continue to work hard (doge).Thanks to the author of Juggernaut and other contributors in the community for their open-source content.Thanks to friends who have always supported and helped Tusan, as well as the first-party dads who have provided financial support through commercial collaborations, allowing Tusan to expand computing power and continue to share open-source.

Files with SHA256: 9c0f2d6a7b08df2d2d61ff24ce55af4c6b9319a00dbd4c643d05b368f3175eb7