HaveallSDXL

Recommendations for generation: CFG Scale: 3For portraits, it’s better to use CFG Scale: 2-3, you can increase it to 4 for more colorful arts, for graphics you can use 3-5.Main formats: 1024x1024, 832x1216, 1216x832The model works best with the DPM++ 3M SDE Exponential sampler.Do not use any Lora, embeddings, aditailers, and other additions. All of this was created on old models and will only interfere with generation on the Haveall SDXL model. You can only use new Lora, which will be trained on Haveall SDXL and only them.SHOWCASE HAVEALL MODEL -> https://haveall.net/ is featured on the model's website. The site shows several hundred examples with images and complete prompts for generation.In the gallery, you can view pictures and prompts. The pictures on the site already have built-in prompts, so you can drag any picture into the PNG INFO tab and get a ready prompt in Automatic or in Forge.For work, I recommend Forge - when installing by default, upscalers will be downloaded and the speed of work is more optimized. I recommend DAT-4 upscaler for high-res fixes (which is built-in by default in Forge).Main advantages of the model:- Very high quality of generated images. Such images will be easily accepted on stock sites, you can create paid covers for writers, musicians, artists. You can process photos, restore photos.- Improved color palette. Improved display of skin, materials, surfaces, and architectural structures.- For training, 5200 very carefully selected images were used. To achieve such quality on one dataset, 4 different models were trained (each with its own parameters), then 4 Lora were created from the models. 4 Lora were combined into one and combined with the model. The training was not on the base model SDXl, but on the HelloWorld 5.0 GPT4V model. This was done to improve the model’s understanding of natural language.- Captions for dataset images were divided into several categories. Part was written by hand, for the caption of a third of the images, GPT chat was used, for part of the dataset, prompts from MJ v.6 were used.- The dataset includes both photos and animated arts, isometry, graphic arts, anime. Different formats and cameras.- Examples of styles can be seen in the gallery.Сайт модели: https://haveall.net/На сайте можно найти много разных дополнений, инструкций и тысячи готовых промптов.Чтобы обновления моделей выходили чаще, меня можно поддержать на бусти: https://boosty.to/artxtopДля связи я в ВК