⬇Read the info below to get the high quality images🙏 (click on show more)⬇
Version Change Log
Tips:There is no significant improvement relationship between all released versions, and each version has its own advantages and disadvantages. Please choose according to your own needs and preferences.
V34.6：International and more diverse character
V30: Comprehensive improvement with significant improvement in realism and generalization.
V29: A natural and round face with a more coordinated body ratio and a more realistic visual texture
V28: Beautiful face and long legs, more aesthetically pleasing visuals
版本说明
所有已发布版本之间并无明显的改善关系，各版本都各有优缺点，请根据自己的需求与喜好选择。
V34.6：国际化且更多样的角色
V30：全面性的改进，在真实度与泛化性有大幅提升
V29：自然圆润脸身体比例更协调，画面质感更真实
V28：脸美腿长，画面更唯美
Introduction / 简介
Most realistic models aim to enhance film grain and blur to achieve a realistic photo effect, but images turning yellow and blurry. The ultimate goal of this model is to maintain high quality while maintaining authenticity.
One of the reasons for creating this model was my desire to have a foundational model that combines aesthetics with realism. I noticed that most models tend to fall into two extremes: some are realistic but lack artistic appeal, while others have beautiful visuals but lack realism. Thus, I developed this model, which, in my opinion, strikes a balance between aesthetics and realism, while being rooted in Eastern aesthetics.
Another reason behind creating this model was to have a highly compatible foundational model. Many existing models tend to be heavily biased towards specific styles, whereas I wanted a model that is truly versatile, allowing me to seamlessly transition between various styles, while relying on the diverse capabilities of the Lora community.
The model only needs to use a simple keyword to draw realistic photos, the whole body image has been greatly improved.
The model purposely reduces the skin blemishes of the young female characters. If you need to create a true sense of blemish, please emphasize it in the prompts.
If you think the model is good, don't forget to give it a five-star rating. Your support is the motivation for me to share.
大部分写实勒模型都是在增强胶片感与模糊来取得类似现实照片的效果，这导致图片往往发黄模糊，而这个模型的最终目标是高质量的同时保持真实性。
做这个模型的一个原因在于我想要一个兼具美学与真实感的基础模型，但大部分模型都是在走两个极端：一部分写实但欠缺美感，另一部分画面唯美但却不够写实。所以，我制作了这个个人认为平衡了美感与真实感的模型，当然这是基于东方审美的基础上。
另外一个原因则是，我想要一个兼容性极强基础模型，但大部分模型都特别偏向于某一风格，而我希望这个模型真的足够基础，让我可以很好的变换各种风格，而其他工作交给社区里种类繁多的Lora。
本模型只需要使用简单的关键词就可以绘出逼真的写实照片，对全身图像进行了大量改进。
模型特意减少了年轻女性人物皮肤瑕疵，如果需要营造真实的瑕疵感请在提示词中强调。
觉得模型不错的话，别忘了给个五星好评~ 你们的支持才是我分享的动力。
About VAE / 关于VAE:
The model is built-in Vae. The light, shadow and color effects are soft. If you like higher contrast images, you can switch directly to VAE or download the non-VAE version to use any VAE you like, or use the followingg: sd-vae-ft-mse-original.
模型已经内置VAE，光影与颜色效果柔和，如果你喜欢更高对比度的图片，你可以直接切换其他VAE使用，或者下载无VAE版本使用其他任何你喜欢的VAE，或使用下面这个sd-vae-ft-mse-original。
Recommend Promts / 推荐提示词:
( The above displays different outcomes with varying "Clip skip" under the same conditions. / 上面展示了在相同条件下不同Clip skip的不同结果 )
Tips:
CLIP Skip：1 or 2 （Choosing 1 leads to results that are more aligned with the prompt and are more accurate. Choosing 2 yields more creativity and can also help reduce overfitting, depending on your preferences and needs.）
You can directly produce images with resolutions of 512*768, or higher, and then Hires Fix.
Orange Color : higher weight ≈ High Saturation & high contrasty
Pink Color = Optional
Option way to less contrasty: To make the image less contrasty you can use LoRA [Detail Tweaker LoRA] in a negative value.
CFG Scale( lower = real): 5 ~11
When used the CFG Scale Fix plugin, you will achieve astonishing images. Please search for usage instructions on your own.
Sampler reconmmend: DPM++ SDE Karras / DPM++ 2M SDE Karras / DPM++ 2M Karras
Draw the whole body image please cooperate with the adetailer plug-in to redraw the face for better results
提示:
Clip Skip：1 或 2 （不同的选择会获得风格不同的结果建议你全部都尝试一下看哪种是你的最爱）
推荐初始分辨率512*768之后进行高清修复放大
橙色 : 提高权重 ≈ 高饱和度 & 高对比度
粉色 = 可选
降低饱和度的其他方法: 可以将 LoRA [Detail Tweaker LoRA] 的权重调成负值。
CFG Scale( lower = real): 5 ~11
配合CFG Scale Fix插件你将得到叹为观止的图片，用法请自行搜索
采样器建议: DPM++ SDE Karras / DPM++ 2M SDE Karras / DPM++ 2M Karras
绘制人物全身图像请配合 adetailer 插件进行脸部重绘以获得更好的效果
Promts sample / 提示词举例:
RAW photo, 1igrl, goddess, white hair, blush, (smile:0.8), upper body, (high detailed skin:1.2), 8k uhd, dslr, soft lighting, high quality, film grain, Fujifilm XT3Negative Promts / 负面提示词:
(deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, sketch, cartoon, drawing, anime, mutated hands and fingers:1.4), (deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, disconnected limbs, mutation, mutated, ugly, disgusting, amputation,BadDream, UnrealisticDream,
Option Negative Embedding / 可选负面Embedding:
Description
No VAE version
FAQ
Comments (10)
It seems like people are confused with 'artistic' and 'realistic' in using SD. They must not be separable if you name your model with 'photo realistic.' If your model is good in overall image qualities and technical aspects, the model MUST generate both artistic and realistic image at the same time WITH PROPER PROMPTS and LORAS. If yours is not, it needs updates and more trial-and-errors because something is wrong. I just wanted to point out that your considering 'artistic' same as 2D is problematic and if you work harder with creating a true realistic model, you will see those are not separable at all.
Different language, as u can see.
@cjs_young oh I c! sorry. good job by the way! :)
这个模型非常棒！收藏了
大佬您好，麻烦问一下，我用了您生成图片的模型，embedding和嵌入词，随机种子和迭代什么的全都是一样的。但是为啥细节方面和动作之类的还是不一样呢。请问我该怎么做才能达到您这种效果。
1、硬件条件不同也会造成无法完全还原例图 2、另外需要注意反向提示词中我用到的embeddings你是否已经下载 3、如果使脸部问题看看是否有启用ADetailer插件
@Tian__ 我后面把缺少的embeddings装了后确实就只剩下一些表情和服饰的区别了。感谢您的指点。顺便再请教一下，您初始生成的时候是选的何种分辨率。变高清的时候是不是用的SD放大呢。还有就是有没有其他的修复方法。因为我自己测试修复以后衣服还是没有那么真实。
@fqmf 初始生成尺寸一般为512*768 然后你可以直接使用高清修复放大2倍（推荐使用DPM++ SDE Karras采样），或者使用图生图放置放大2倍，如果想要更大尺寸的图片你可以试着使用Controlnet的tile模型加上 Ultimate SD upscale 或者 MultiDiffusion 的"分块 VAE"功能来进行分块放大，你可以详细阅读模型简介中的内容，希望对你有所帮助。
@Tian__ 非常感谢大佬的帮助！
新人，感谢指点。