这是一个超逼真的视觉模型，你可以使用该模型生成媲美真实世界照片的图像,
请勿使用该模型生成图像用于违法犯罪，请遵守各使用者的地区法律,
该模型在2*4090上训练500小时，训练集的制作时间长达30天,
作者仅在webui中进行模型测试，在comfyui下的效果并未实测,
由于对照片真实感的特化，该模型并不具备除照片以外的任何泛用性,
由于一些特殊的原因，使用该模型请控制CFG为3.5-4.5（例图为4），否则会出现图片崩坏
该模型在特写镜头下进行高清重绘，会出现多眼睛，五官扭曲的情况,该情况是因为训练分辨率不足以支撑高清重绘导致的，请适当降低放大倍率或者重绘幅度,
推荐的放大算法：Lanczos（该算法下可避免gan放大算法出现的涂抹感）
推荐重绘幅度：特写0.1，半身0.25，全身0.3，场景可固定0.3
在超出1024出图分辨率下，请配合Kohya Hires.fix插件进行使用，可有效避免出现重复元素和多余肢体,
请使用自然语言描述图片，以获得更好的效果
不支持任何lora,loha,locon,LyCORIS(因为是用lokr调整的，）
可以配合ttp（汤团猪）的https://civitai.com/models/330313/ttplanetsdxlcontrolnettilerealisticv1
进行图片放大，重绘等操作（请勿用于非法用途）
感谢该模型制作中各位朋友的帮助：
https://civitai.com/user/MIAOKA
https://civitai.com/user/ttplanet
该模型暂无下一步更新计划（随我高兴），暂不公开训练集与训练方法（略略略就不给）,如有商业需求，请与我联系
Bilibili：https://space.bilibili.com/164860411
祝你们玩的开心，愿世界和平
This is a super realistic visual model that you can use to generate images comparable to real-world photos.
Do not use this model to generate images for illegal activities. Please comply with the laws of each user's region.
The model was trained on 2 * 4090 for 500 hours, and the production time of the training set was as long as 30 days.
The author only conducted model testing in webui, and the effect was not measured in comfyui.
Due to the specialization of photo realism, this model does not have any universality other than photos.
Due to some special reasons, when using this model, please control the CFG to be 3.5-4.5 (as shown in Figure 4), otherwise image collapse may occur
This model undergoes high-definition redrawing in close-up shots, resulting in multiple eyes and distorted facial features. This situation is caused by insufficient training resolution to support high-definition redrawing. Please reduce the magnification or redraw amplitude appropriately.
Recommended amplification algorithm: Lanczos (this algorithm can avoid the smudging sensation caused by gan amplification algorithm)
Recommended redraw range: close-up 0.1, half body 0.25, full body 0.3, scene can be fixed by 0.3
When the image resolution exceeds 1024, please use it in conjunction with the Kohya Hires.fix plugin to effectively avoid duplicate elements and unnecessary limbs.
Please use natural language to describe the image for better results
No support for Lora, Loha, Local, LyCORIS (as it was adjusted using Lokr)
Can be paired with TTP https://civitai.com/models/330313/ttplanetsdxlcontrolnettilerealisticv1
Perform image enlargement, redrawing, and other operations (do not use for illegal purposes)
Thank you for your help in creating this model
https://civitai.com/user/MIAOKA
https://civitai.com/user/ttplanet
There is currently no plan for the next update of this model (as I please), and the training set and methods are not publicly available. If you have any business needs, please contact me
Email: [email protected]
Bilibili: https://space.bilibili.com/164860411
Wishing you all a happy time and wishing for world peace
描述:
训练词语:
名称: white_v10.safetensors
大小 (KB): 6938818
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
