Guilingao – V3

Guilingao - V3

Run Model 使用GRAVITI Diffus 关于GRAVITI Diffus
Model Name Guilingao
Model Type Checkpoint
Base Model SD 1.5
Version Name V3
File Name guilingao_v3.safetensors
SHA256 efb8a84c7b
NSFW False
Trigger Words
Tags anime, character
Metadata fp = fp16
size = full
format = SafeTensor
Size 5.57 GB
Version Description

This is the first version that satisfies me.

这是第一个令我满意的版本。

Source https://civitai.com/models/11427
Author Arzayou

Based on a fusion of dreambox-mix, cetusMix_cetusVersion2 and Counterfeit-V2.5.

The EasyNegative program is required to restore the images in the example. If this is the first time you have heard of EasyNegative you can get a brief overview and download it here https://huggingface.co/datasets/gsdf/EasyNegative

Please ensure that the Hires. fix function is used when generating images, my usual settings are Denoising strength: 0.6, Hires upscale: 2, Hires steps: 12-20, Hires upscaler: Latent. With Steps: 20-24, Sampler: DPM++ 2M Karras, CFG scale: 8-10 are used for generating 576×768 images. I usually use “((masterpiece, best quality)),” as the start of a positive prompt and “EasyNegative,” as the start of a negative prompt.

It was originally made with the intention of being able to draw fantasy medieval (or some kind of JRPG setting) in an alternative setting to my own satisfaction. As this is a spur-of-the-moment piece to satisfy personal preference, it has not been fully tested and may not be suitable for pictures in certain settings. In addition, my experience of using it shows that its positive prompt has problems with erratic image generation at less than 40 and more than 200. I have not been able to carry out similar tests with the negative prompt due to my use of EasyNegative (as the use of EasyNegative puts the negative prompt at 20-50 on a regular basis). If this is the first time you have heard of EasyNegative you can get a brief overview and download it here https://huggingface.co/datasets/gsdf/EasyNegative

I would be honoured if it produced wallpapers or avatars to your satisfaction.

Please do not use it commercially or in a way that might infringe on the portrait rights of others.

(Based on DeepL Chinese to English translation, which may be inaccurate.)

基于dreambox-mix、cetusMix_cetusVersion2和Counterfeit-V2.5的融合。

还原示例中的图片需要EasyNegative程序。如果第一次听说EasyNegative你可以在此简要了解并下载它https://huggingface.co/datasets/gsdf/EasyNegative

请确保Hires. fix功能在生成图像时被使用。我通常的设置为Denoising strength: 0.6, Hires upscale: 2, Hires steps: 12, Hires upscaler: Latent。配合Steps: 22, Sampler: DPM++ 2M Karras, CFG scale: 10的参数对生成576×768图片时使用。我通常使用“((masterpiece, best quality)),”作为正prompt的开始,用“EasyNegative,”作为负prompt的开始。

由于最初的目的是能够画出另自己满意背景下幻想中世纪(或是说某类JRPG背景,像是“为美好的世界献上祝福”那类的世界观)的作画而进行制作。由于仅是一时兴起为满足个人喜好的作品,它没有进行充分的测试,可能并不适应某些背景下的图片。此外,个人的使用经验显示其正prompt在小于40和大于200的情况下会出现图片生成不稳定的问题,而负 prompt由于个人使用EasyNegative的原因未能进行过类似测试(因为EasyNegative的运用使负prompt经常在20-50)。如果第一次听说EasyNegative你可以在此简要了解并下载它https://huggingface.co/datasets/gsdf/EasyNegative

(我称呼这个模型为龟苓膏…就是吃的那个龟苓膏,因为我融合它的时候在吃那玩意。)

如果它画出令你满意的壁纸或是头像,我感到很荣幸。

请不要使用它从事商业行为或是可能侵害他人肖像权的行为。

查看更多

© 版权声明

相关文章