LEOSAM HelloWorld | SDXL Realism Checkpoint

Copyright Statement:

HelloWorld series model (hereinafter referred to as "this model") is developed by me (hereinafter referred to as "the owner") with the support of LiblibAI platform and exclusively launched on the LiblibAI platform as the XL large model. Except for the LiblibAI platform and Civitai platform, any individual or entity who publishes this model on other platforms without the authorization of the owner.

The owner authorizes individuals or institutions to use the images generated by this model for non-commercial educational or informational purposes for free, and:

- Comply with relevant laws and regulations, and do not infringe upon the legitimate rights and interests of this model or any third parties.

- When using the images, it is necessary to indicate the image source as "Generated by LEOSAM's HelloWorld XL model".

For commercial purposes, a commercial authorization agreement must be signed with the owner in advance. For commercial authorization and model customization matters, please contact the owner through the information on the homepage of LiblibAI platform.

The training and open source sharing of the SDXL model is a time-consuming and challenging work. The owner will continue to provide updates of the HelloWorld model for individual players for free as an expression of support and gratitude to the community open source contributors. The paid cooperation of commercial users is an important driving force for the development and continuous improvement of this model. Thank you for the understanding and support of every user.

Please note that any unauthorized use may violate relevant laws and regulations, and may bear legal responsibilities. The final interpretation right of this statement belongs to the owner and is subject to the relevant laws and regulations of the People's Republic of China.

June 13, 2024 HelloWorld Version 7.0 Update

One-sentence Update Summary: HelloWorld 7.0 is an iterative optimization version with the best limb performance in the entire series, and the richness of concept range and detail have also been further enhanced.

Update Details:

1. By adding negative training images, strengthening individual posture training, and optimizing the clip model, the correct rate of the model's limbs and hands has been improved compared to the previous version. Recommended negative prompt words include: "bad hand, bad anatomy, worst quality, ai generated images, low quality, average quality"

2. Extracted and fine-tuned LoRA from the official SPO model and integrated it into HelloWorld 7.0. SPO is a further improvement of the DPO method, and the SPO base model performs better than the DPO XL base model and the original SDXL base model. SPO LoRA can enhance picture details, contrast, and image beautification effects. Thanks to the technical team behind SPO for this.

3. Continuously expanding the concept range of the training set but refining the training set for optimization (fine-tuning is too expensive, and it's been difficult to rent H800 recently, and the time cost is also high). The current training set contains a total of 20,821 images, and the distribution of training set resolutions is as follows, recommend using some of the resolutions with higher quantity for image output:

(832, 1248) - Count: 7128

(896, 1152) - Count: 6250

(1248, 832) - Count: 2402

(1024, 1024) - Count: 1639

(1360, 768) - Count: 928

(1152, 896) - Count: 870

(768, 1360) - Count: 432

(960, 1088) - Count: 506

(992, 1056) - Count: 162

(1088, 960) - Count: 140

(704, 1472) - Count: 120

(1056, 992) - Count: 122

(1472, 704) - Count: 115

(1632, 640) - Count: 75

(640, 1632) - Count: 12

4. GPT4O was used to re-label all datasets, using a structured labeling method this time, with the specific structure being: "A summary description in one sentence + multiple image element tags + inspired by XXX + aesthetic quality description words", where aesthetic quality description words include worst quality, low quality, average quality, best quality, and masterpiece, in five levels. A typical label is as follows:

conceptual art featuring a human hand wrapped in red and beige ribbons, isolated against a plain, light background, realistic style, minimalist color scheme, smooth textures, elongated and surreal aesthetic, inspired by salvador dalí's surrealist works, masterpiece

The "High-Frequency Trigger Word List" for HelloWorld 7.0 and the "High-Frequency Art Style List" related to Inspired by XXX will only be provided to commercial authorized users. For partners who have purchased authorization for Helloworld XL series models in the past and might have missed out, please contact me for a free update. Players can refer to the high-frequency label word document for HelloWorld 6.0. I have also provided 180 high-quality outputs of HelloWorld 7.0 in the return image area for everyone's reference. Model creation is not easy, thank you for the understanding and support of all players.

The permanent authorization fee for the entire Helloworld XL series (including versions 1.0 to 7.0, inpainting version, and lightning version) is 6,000 yuan for small team authorization (producing less than 5,000 images per month) and 20,000 yuan for large client authorization (producing more than 5,000 images per month). Authorization agreements/contracts can be signed, and value-added tax invoices are provided. In addition, I provide technical consulting services related to model training, with a consulting fee of 2,000 yuan per hour (comparable to consulting fees in the lawyer industry). The fees for model customization services need to be discussed based on requirements and workload.

It's time-consuming for both parties to inquire one by one, so I'm posting it clearly. If you are interested, please add my WeChat 13439222668 for commercial authorization consultations.