XXMix_9realistic

Version 3.0 Update: The XXMix_4_v2893 model in the image information is the v30 version. The image was generated before the model was renamed. 1. I am pleased to introduce the 3.0 version of the XXMix_9 series to everyone. This version has significantly enhanced universality and support for LORA, and the performance of the hands is noticeably better than the previous versions. The 3.0 version is a mix of versions v2.5, v2.6, and XXMix_4. I tried to generate images withou* ***ng any Embedding model as I found some Embedding models to be ambiguous. You can't tell which Embedding models are working and which are not, making it impossible to restore the legend. 2. If you want to restore the legend, please install the following extension first: Adetailer: It is recommended to always be enabled when generating images (if you want to restore the legend by more than 90%, then check if adetailer is enabled in the image information). Note: Adetailer needs to download some models from https://huggingface.co/Bingsu/adetailer/tree/main. Here is the model address. Model storage path: webui/models/adetailer. If the adetailer folder is not found, create one. Adetailer video tutorial, here is Nausicaa's tutorial I posted on Bilibili https://www.bilibili.com/video/BV13s4y197cK/?spm_id_from=333.337.search-card.all.click&vd_source=8c2a86ab53df9752cdd75679f58834a6 CFG Scale Fix: Enable it when needed to improve the texture of the image. I use the default value around 7. If it is daytime, the CFG value can be slightly larger; at night, the value should be smaller. (After my testing, I couldn't restore the legend 100% on the second computer. I still don't know the reason. It might be an issue with the CFG Scale Fix extension, or it might be a problem with the block VAE, but basically you can restore it to about 90%-95%, and the impact on the image is not significant.) 3. Personal tips for generating images: If you have obtained an image that you really like but there are serious problems with the hands or other parts, and you don't want to use local repaint to make modifications, then I suggest that after fixing the seed, adjust the CFG value slightly (try adjusting by 0.2-0.3 each time), this way you may get a perfect image while preserving the original. 4. Here, I would also like to share a suffix sentence that I like, it can make the image more expressive. Of course, you can modify the words in the sentence as needed to fine-tune the visual style. These words include: fantasy, high contrast, ink strokes, explosions, overexposure, purple and red tone impression, abstract, ((watercolor painting by John Berkey and Jeremy Mann)) brush strokes, negative space. 5. Finally, if you like the updates to this model, hope you can share the generated beautiful images. ------------------------------------------------------------------------------------------------------- Version 2.6 Update: This is a minor update, optimizing the face shape, improving hand issues, but mainly gashapon. Major issues will be addressed in the next version. ------------------------------------------------------------------------------------------------------- Someone commented that the model is basically unusable. Please check if your settings are the same as the legend's settings, except for the full-body image, which may have IMG2IMG zooming operations, all other images are straight out. Please make sure your settings are the same as in the image!! Version 2.5 has better light and shadow effects, supports more fantasy content, adjusts the basic face shape, and shows scenes better. Adjusting the value of CFG Scale appropriately can give you a unique effect. Version 2 Update: 1. Changed the default face. 2. Better light and shadow effects. 3. Incorporated more models, will continue to iterate in the future. 4. If you want better skin texture, do not enable high-definition repair when generating images, instead pass the generated image through the image-to-image mode, use Tiled Diffusion to upscale, to get a more realistic skin texture. 5. I will continue to update and iterate on this huge model, hoping to continuously enrich the content and make the model more interesting. Thank you for your support! ---------------------------------------------------------------- Recommended Parameters: Sampler: DPM++ 2M Karras alt Karras or DPM++ SDE Karras Steps: 20~40 Hires upscaler: 4x-UltraSharp Hires upscale: 2 Hires steps: 15 Denoising strength: 0.2~0.5 CFG scale: 6-8 clip skip 2