[Checkpoint] Fantexi_v0.9Beta
模型介绍：mix of many models, VAE is baked，good at NSFW
很多模型的混合，vae已经烘焙，擅长nsfw
setting:
▣ Denoising strength: 0.55,
▣ Clip skip: 2,
▣ ENSD: 31337,
▣ Hires upscale: 4,
▣ Hires steps: 15,
▣ Hires upscaler: R-ESRGAN 4x+,
▣ Eta: 0.68
2023.3.19 update
添加了两个版本，0.8a更真实，0.8c更动漫画
Two additional versions were added, 0.8a being more realistic and 0.8c being more anime
2023.6.13 update
更新测试版本0.9，此版本非常不稳定，很考验玩家的负面关键词使用熟练度，达到示例图效果需要下载所需lora，必备面部修复插件和动态阈值插件，此版本不再烘焙VAE，但我个人建议还是使用vae-ft-mse-840000-ema-pruned,示例图全部由此VAE生成，经测试此版本容易出现手脚崩坏和服装单一问题，建议使用相关lora配合使用，所有示例图片可以直接拖入SD图像浏览器查看相关参数
Update test version 0.9, this version is very unstable, it is a test of the player's proficiency in using negative keywords, to achieve the effect of the example image, you need to download the required lora, necessary face repair plug-ins and dynamic threshold plug-ins, this version no longer bakes VAE, but Personally, I recommend using vae-ft-mse-840000-ema-pruned. All the example pictures are generated by this VAE. After testing, this version is prone to hand and foot collapse and clothing oversimulation. It is recommended to use related lora to cooperate with it. All picture parameters Can be directly dragged into the SD image browser to view related parameters
关于图像模糊问题
请使用controlnet进行分块放大修复，如果这还帮不到你你可以借助外援继续放大，电脑不太好的玩家建议使用Topaz和Upscayl进行放大，Upscayl目前是免费的而且速度很快，相关使用教程可以在YOUTUBE和BILIBILI上找到，有很多的视频博主讲解了contronlnet的使用方法，另外图像设置也尽量按照示例图参数不要随意改动，图像采样一般只用DPM++SDE，超级分辨率放大算法请使用4x-UltraSharp，当然其它的算法也可以自行尝试
About image blur
Please use controlnet to zoom in and repair the blocks. If this still doesn’t help you, you can continue to zoom in with the aid of foreign aid. Players with poor computers are recommended to use Topaz and Upscayl for zooming in. Upscayl is currently free and very fast. Related tutorials It can be found on YOUTUBE and BILIBILI. There are many video bloggers explaining how to use contronlnet. In addition, the image settings should also follow the example image parameters as much as possible and do not change at will. Generally, only DPM++SDE is used for image sampling. Super resolution zoom-in algorithm please Use 4x-UltraSharp, of course other algorithms can also be tried by yourself
写了这么长简介还不看，在评论区问的一律拖出去斩首
If you don’t read such a long introduction, anyone who asks in the comment area will be dragged out and beheaded
~~~~~~~~~~~~~~~~~~~~~~~o(*￣︶￣*)o~~~~~~~~~~~~~~~~~~~~~~~
示例提示词和参数：
Prompt: ((magical girl)), ((gorgeous starry sky background)), ((ultra-detailed)), ((cinematic lighting)), dynamic angle, (huge breast:1.2),floating, finely detailed, (glitter:1.2), (sparkle:1.2), (shine:1.2), classic, (painting:1.1), (sketch:1.1), (best quality), (masterpiece:1.2), (solo), beautiful detailed face, colorful hair, (long hair:1.2), (floating hair:1.2), (magical girl outfit), detailed outfit, (magical effects:1.3), (flowing skirt:1.2), (cute boots:1.1), (transparent fabric:1.1), (twinkle:1.2), (glow:1.2), (radiance:1.2), (flicker:1.2), (dazzle:1.2)
Negative prompt: (three hands:1.1),(three legs:1.1),more than two hands,more than two legs,(easynegative, bad-artist-anime, bad-artist bad-hands-5, badhandv4, bad_prompt_version2, ng_deepnegative_v1_75t,bad-picture-chill-75v,,verybadimagenegative_v1.3),sketch by bad-artist), (loos leg:2),((((ugly)))), (((duplicate))), ((morbid)), ((mutilated)), , extra fingers, mutated hands ,((poorly drawn hands)), ((poorly drawn face)), (((mutation))), (((deformed))), ((ugly)), blurry, ((bad anatomy)), (((bad proportions))), ((extra limbs)), cloned face, (((disfigured))). out of frame, ugly, extra limbs, (bad anatomy), gross proportions, (malformed limbs), ((missing arms)), ((missing legs)), (((extra arms))), (((extra legs))), mutated hands, (fused fingers), (too many fingers), (((long neck))), signature, text, facial hair, loli, chibi,disabled body,DeepNegative
Steps: 20, CFG scale: 8.5, Sampler: DPM++ SDE, Seed: 3918891357, Eta: 0.67, Size: 896x1496, Model: fantexi_v0.9, Version: v1.3.1, Model hash: ba4497f5a6, AddNet Enabled: True, AddNet Model 1: PerfectFullBreasts-fCV3(9572b9135c1a), AddNet Model 2: uzaki tsuki 3(e87f3975aaa7), AddNet Module 1: LoRA, AddNet Module 2: LoRA, AddNet Weight A 1: 0.35, AddNet Weight A 2: 0.4, AddNet Weight B 1: 0.35, AddNet Weight B 2: 0.4, Denoising strength: 0.15, Clip skip: 2
页:
[1]