Any questions or corporations? Find me at Discord or by Wechat: ghostinshell10. Thanks
You can run GhostMix on the cloud at Mage & SinkIn.ai:
My model is now supported to create a short-form animation with the new "Animate" feature on Mage!
GhostMix V2 at: https://www.mage.space/model/ghostmix-v2
GhostXL at: https://www.mage.space/model/ghostxl_v10BakedVAE
SinkIn.ai GhostMix model at: https://sinkin.ai/m/DY5rYnx
IMPORTANT MATTERS(重要事项)
I think compacity is the most important thing of a checkpoint, that's why I don't merge ANY LORA in GhostMix. Checkpoint solves the CAN DO problem and Lora solves the DO IT RIGHT problem. （我认为checkpoint最终要的是兼容性，所以我没有融任何lora进checkpoint，checkpoint应该解决的是做的到的问题，而lora解决的是做的对的问题）
Highres-Fix is A Must! Highres-Fix: 2x, denoising:0.4-0.5 or 1.5x, denoising:0.5-0.65. (一定要做高清修复! 高清修复: 2倍, 重绘幅度:0.4-0.5 或 1.5倍, 重绘幅度:0.5-0.65)
Make Sure you are in the right CLIP if you want to replicate my job, some themes are CLIP=1,while others are CLIP=2.Suggest download the image and put it into PNG info to check the setting (如果想要复现,确保CLIP值要对! CLIP1和CLIP2要对!建议把图下下来然后放到PNG信息里面去查设置)
Most Prompts in Previous Version of GhostMix can produce similiar result in New Version of GhostMix (之前用的大多数Prompts在新版本也可以生成相似的结果)
Textual Inversion&VAE: ng_deepnegative_v1_75t and easynegative ,don't use Bad-Hand V4 & V5！（用 ng_deepnegative_v1_75t和easynegative,别用BadHandV4,V5)
Sampler Suggest : DPM++ series , Steps: 20-30, CFG:5-7(7 is best)（采样方法建议 DPM++系列 , 步数20-30, CFG:5-7(7最好) ）
Suggest resolution: 512,768! mechanical girl theme is very sensitive to the resolution, not suggest make the aspect ratio too low.(建议分辨率:512,768! 机械少女主题对分辨率设置非常敏感，不建议设太低的宽高比)
If you want to support me, please buy me a coffee : https://ko-fi.com/ghostshell
国内支付宝、微信用户可以通过爱发电给我买杯咖啡：https://afdian.net/a/ghostmix
I create the world's first checkpoint review framework GhostReview, by using LAION Aesthetics, Clipscore, HPS, StyleLoss to measure ckpts quantitatively in Image Quality,Style Compatibility and LoRA Compatibility. If you want to see the detail, please check the article link below.
GhostReview Github Project: https://github.com/drnighthan/GhostReview
2023.5.21 GhostMix-V2.0 (fp16 pruned ver replaced)
UPDATE DETAIL(中文更新说明在下面)
Hello everyone, this is Ghost_Shell, the creator. The GhostMix-V2.0 significantly improves the realism of faces and also greatly increases the good image rate. In my tests at 512,768 resolution, the good image rate of the Prompts I used before was above 50%. It is more user-friendly. During making the GhostMix-V2.0, I adjusted 47 versions of the model and finally chose one of them.
大家好，这里是作者Ghost_Shell。这次GhostMix-V2.0大幅提升了脸的真实性，也大幅提升了良图率，在我测的512,768分辨率之下，之前用的Prompts良图率都在50%以上，对用户更加友好。这次在测试中一共调了47个版本的模型，最终选了一个。
Other Words I want to say(题外话):
To be honest, this may be the last version of GhostMix.On one hand, it is really inefficient to use 3060ti to make models. In the past two weeks, I have almost no free time except for making models and testing them. On the other hand, I temporarily feel that this model is almost at its limit and the space for improvement is really not high. I hope you like it. If you like the model, I hope you can post your images to Civitai. Many of the prompts I tested for GhostMix-V2.0 are from your posts, which is really important for me to test the model. If you can give it a 5-star rating, that would be great. If you are willing to support my work, please click: https://ko-fi.com/ghostshell. My goal is to buy a 4070 and work more efficiently on making models. Using a 3060ti to make models is really inefficient and it basically cannot be used to test high resolution images.
说实话，这可能是GhostMix的最后一个模型，一方面3060ti去做模型真的效率太低了…最近两个星期基本没有空闲时间，除了做模型，就是测模型。另外一方面，我暂时觉得这个模型近乎极限，能提升的空间确实不高了，希望大家喜欢。如果大家喜欢模型，希望大家能post自己的作品到Civitai，这次测试的很多Prompts就是从你们post里面来的，这对我测试模型真的很关键，如果能5星评价就更好。如果愿意支持我的工作，请点击：https://ko-fi.com/ghostshell。 我的目标就希望能换一块4070，更高效的去做模型，3060ti真的测模型效率太低了，而且基本没法测更高分辨率的图片。
2023.5.1 GhostMix-V1.2 (fp16 version uploaded)
UPDATE DETAIL(中文更新说明在下面)
THIS IS NOT A 3D MODEL! THIS IS NOT A 3D MODEL! THIS IS NOT A 3D MODEL!
If you like my model，please give me 5 Stars ,it will encourage me a lot. Thanks!
Color Problem ：Check VAE is kl-f8-anime2 ？颜色问题：查VAE是否为kl-f8-anime2？
GhostMix V1.2 is an absolutely astonishing model, and I think it is the strongest 2.5D model in Civitai right now. I think a update of the model should improve the model’s compatibility, good image rate and image details given the main structure of 90% generated images doesn’t change. So I use layer combination to combine models layer by layer. And I got this model after 9 different versions of “the final version of GhostMix V1.2”,LOL.This version of GhostMix V1.2 is a balance in terms of compatibility, good image rate and image details, although sometimes using the same Promts of GhostMix V1.1, may comes up a different result.But this doesn’t happen too much.
PS: The first image of GhostMix V1.2 is a nod to Mamoru Oshii’s version of Ghost in Shell in 1995, and my name Ghost_Shell is also a nod to this movie.
中文更新说明
GhostMix V1.2是绝对让你惊艳的模型，也是自己认为现在最强的2.5D模型。我认为模型的更新应该是基于现有的画面整体不大变的前提下，提高模型的成图率，兼容性和画面细节。所以我采用了分层融合，一共做了9个版本的“GhostMix V1.2最终版本”，最终得到了现在这个版本的GhostMix V1.2。这个版本的GhostMix V1.2在兼容性，成图率和画面细节表现比较平衡，虽然有时候会出现同样的Promts，出图改变的情况，但实测不多。
PS：GhostMixV1.2的头图是致敬1995押井守版Ghost in Shell 攻壳机动队的生成的，Ghost in Shell也是作者名字的由来。
2023.3.11 GhostMix-V1.0
Introduction(中文简介在下面)
First of all , I want to thank all the people who use this Checkpoint. And this is my first Checkpoint.All the sample image can be reproduced. This Checkpoint works well on both SFW and NSFW.THE NSFW PART IS VERY GOOD!!!
I uploaded this Checkpoint yesterday and from yesterday to today , I am still trying all the possibility of this Checkpoint. So if you try the model and find some good promts , I hope you can upload it and share with me, it will help me for the next version of GhostMix, Thank you agian!
Recommend Some Promts:
Fractal Art（highly recommend，awsome）
(masterpiece, top quality, best quality, official art, beautiful and aesthetic:1.2), (1girl:1.3), (fractal art:1.3),Color Art
(flat color:1.3),(colorful:1.3),(masterpiece:1.2), best quality, masterpiece, original, extremely detailed wallpaper, looking at viewer,1girl,solo,floating colorful waterVAE&Textual Inversion:
VAE: kl-f8-anime2 or vae-ft-mse-840000-ema-pruned(anime suggest: kl-f8-anime2)
Textual Inversion: ng_deepnegative_v1_75t, easynegative
About Image Reproduction:
Some user said that they can not reproduce my result.Maybe the setting goes wrong.I just reproduce my cover image. If you want to reproduce my result, Checkpoint Model，Postive Promt，Negative Promt，Textual Inversion，Sampler Steps，Sampler，CFG，Resolution，Seed , ALL the things should be the SAME! Then be careful if you open controlnet or lora ,don't make them influnence your result.
中文简介
首先感谢每一个使用这个Checkpoint的人，这是我融的第一个Checkpoint。所有样图自测都可以复现。SFW和NSFW的图都挺漂亮的，NSFW的图非常棒。这个模型昨天上传到今天，我一直在尝试这个模型的可能性。希望使用这个Checkpoint的朋友们，如果你试了，觉得有不错Promts，欢迎上传图分享，让我也了解一下这个Checkpoint的可能性，也可以帮助我调下一个版本GhostMix，再次感谢。
推荐Promts:
分型艺术（超级推荐，必出好图）
(masterpiece, top quality, best quality, official art, beautiful and aesthetic:1.2), (1girl:1.3), (fractal art:1.3),色彩艺术
(flat color:1.3),(colorful:1.3),(masterpiece:1.2), best quality, masterpiece, original, extremely detailed wallpaper, looking at viewer,1girl,solo,floating colorful waterVAE&Textual Inversion:
VAE:kl-f8-anime2或者vae-ft-mse-840000-ema-pruned(动画风建议kl-f8-anime2)
Textual Inversion:ng_deepnegative_v1_75t,easynegative
关于图片复现:
有同学好像没法复现我的图，可能是设置有点问题，刚刚我才把头图给复现了。注意几个点：Checkpoint 模型，正向Promt，反向Promt，Textual Inversion，迭代步数，迭代方法，CFG，分辨率，随机种子都必须一模一样！然后复现的话，要把controlnet和lora关掉，怕影响复现。
Description
FAQ
Comments (165)
Hi,everyone.I am the creator of this model, if you have any question or suggestion please tell me here. I will try my best to improve it in the next version.If you like my model, please give me 5 stars review , it encourage me a lot. And you can upload the pics you made,it will help me for testing the next version of model. If you don't know how to use GhostMix, you should check the Instruction part.
Could you upload a noVAE version too? pls :)
@Cala_Cunca ok , I will do it tonight
I am trying to use Hires.fix to upscale a mechanical girl rendering. At 512x768, her face renders fine, but when it tries to upscale, it mangles her face (just turns it into a strange meat-tube).
@limblessgirl4172 maybe set the denoising a little bit lower? Or maybe change to another image....My personally recommendation of Highres-Fix is :2x, denoising:0.4-0.5 or 1.5x, denoising:0.5-0.65
@Ghost_Shell which upscaler do you perfer? I was using "R-ESRGAN 4x+ Anime6b"
@limblessgirl4172 For realistic image , I suggest to use 4X Ultrasharp , you can google it. Otherwise, "R-ESRGAN 4x+ Anime6b" is a good choice.
Hi, thank you so much for making this model. I've been using it for a week and it's my absolute favourite and has given me the best results. However, I've been struggling to make characters with Animal heads/helmets/armour. I've seen in your gallery that you've successfully made animal helmets by using the name as a simple prompt (Chinese dragon) or called it a mask (helmet oni evil mask). I've copied most of your negative prompts and followed your instructions. Namely, I've been trying to create "rat angel knights/samurai with wings" but the model seems to struggle with it a lot and the head gets very messy or looks more like a mouse. I've circumvented the messy heads by using the Cyberhelmet LORA but maybe that's why I do not get proper animal heads. There have been very few occasions where the model showed the ability to create actual rat-shaped helmets. I'll upload my attempts in my review. I'd love to hear your insight and maybe future versions make animal heads easier.
@qng211 Thank you for your advice. This is my fault , I just didn't test image like that. But I think it can solve by lora. If I understand you right , maybe just check this lora , https://civitai.com/models/60869/gpts3-fancy-animal-portraits-or-lazy-lora-or-extended-lora?modelVersionId=65352.
大家好，我是作者，如果有任何问题，欢迎来这里留言，在自己能力范围内会尽力给各位解答，并尽力在下个版本帮大家解决。如果大家觉得模型效果不错，请帮忙给模型五星好评，这对我有大的激励作用，感谢。欢迎大家上传自己做的图片，测试模型需要各种各样的Promts，分享Promts可以帮助我做出更好的模型，再次感谢。如果你不会用这个模型，请看模型简介部分。
Thank you for your hard work. Your model is just gorgeous!
Thank you !
hey thanks man... your model is i like most..
Thank you for your love
can someone help me with this error plzzzz :(
loading Lora C:\Users\Ari\stable-diffusion-webui\models\Lora\ghostmix_v20Bakedvae.safetensors: ValueError
Traceback (most recent call last):
File "C:\Users\Ari\stable-diffusion-webui\extensions-builtin\Lora\lora.py", line 222, in load_loras
lora = load_lora(name, lora_on_disk.filename)
File "C:\Users\Ari\stable-diffusion-webui\extensions-builtin\Lora\lora.py", line 144, in load_lora
key_diffusers_without_lora_parts, lora_key = key_diffusers.split(".", 1)
ValueError: not enough values to unpack (expected 2, got 1)
This is not a lora, this is a checkpoint, you should put in into the model/stablediffusion folder
请问大佬会不会出2.0不带VAE的版本？
this is absolutely one of my favorite models!!! thanks so much!
Thank you for your love
Really impressive. Side by sides with the last version are pretty incredible.
Thank you for using the model
3060ti做的模型吗，很强，加油大佬
是的,谢谢,全靠肝[笑哭]
太惊艳了,沃德天
毕竟做了两个星期，测了47个版本，不惊艳点，对不起我的时间
Why do you say don't use the bad hands textual inversion?
Because the bad hands textual inversion will highly influence the quality of image. It is not only my model, I talk with other generator, they have the same feeling.
@Ghost_Shell Okay thank you :)
请问大佬怎么解决出图过于暴露的问题？
这个我也没有很好的办法，先把NSFW调到2，然后看看正面prompts那里加上衣服的关键词？
请问怎么出暴露的图
这种问题一般不解决
这怎么能是问题呢？
Amazing Work!! This model is a lot of fun to use. I think the 4060ti might be perfect for budget AI training. You could maybe even scale to 2 of them.
Edit:4060ti 8gb is so slow lol. Hope the 16gb is at least a bit better.
Thank you! Maybe you mean the 16G version of 4060ti. But according to the 4060ti 8G review, the whole performance of 4060ti 8G is almost the same as 3060ti G6X. So I think 4060ti 16G may not be a good choice either. I will just wait the review of 4060ti 16G.
@Ghost_Shell Yea, really sorry for suggesting it. I only thought of the high VRAM and not the crap performance lol.
Im using 3080ti with 12 gb vram. It is sufficient for me to run most of stable diffusion at 768x768 but i can tell it is abit slower to generate than 512x512
@yuto303 Thank you for telling me. That's why I think 4070 is a good choice.
太神了,复现率100%,这稳定度太厉害了.
谢谢好评!能复现就好~
请问有支付宝地址吗？想支持但paypal不收中国开头的Visa卡。
非常感谢支持！我想收益全透明，定时公开给所有人看，所以还是走平台吧，刚刚开通了爱发电，再次感谢您。Ghost_Shell正在创作Stable Diffusion模型GhostMix的制作者 | 爱发电 (afdian.net)
@Ghost_Shell 加油呀
@2325190416371 谢谢鼓励
最喜欢的用的模型之一，已经在爱发电捐助，加油！
非常感谢！谢谢！
Your model is seriously impressive! I'd love to share some of the amazing images it generates on my TikTok. Can you let me know if it's okay with you and if I need to mention your model? Thanks!
Thank you ! It's fine to share the images for any non-commercial useage. If you can also mention my model, I will be very happy about that. I want more people to use my model.
Hey. What is your Tik Tok!? Would love to see too!?!??!
Thank you for publishing your wonderful model, please enlighten me on one point. If I merge GhostMix into a model by merging, can I change all 6 permissions are OK for that model? Or, do the following two items have to be inherited?
・Share merges using this model
・Have different permissions when sharing merges
Or is there a way to make all 6 permissions OK for a model that has merged GhostMix?
Hi,thank you for using my model. I just change my policy , I think "Share merges using this model & Have different permissions when sharing merges" is ok. But just credit me if you do so.
@Ghost_Shell Understood. Thank you for your consideration. I don't think I will merge this time. But this model is really great and I cheer you.
@Ghost_Shell I have made a small donation via Ko-fi. Good luck with your development!
@AIbijo Thank you for your support !
@Ghost_Shellhhh
Something to keep in mind that you can always rent out cloud computing to do your training as well. You can rent out a a100 for less then a dollar an hour, cost efficiency in your short term might be able to help with your work flow and greatly reduce your training time, while also still being able to actually play games and what not on your computer while doing so.
Thank you for your advice. GhostMix is a mix model , most of the time is at testing the new merge.But usually the storage of cloud is too small. And it is not convenient to upload many test models (2.2G for one model) to test.
I am trying to convert safetensors to diffusers style, but it fails:
python scripts/convert_original_stable_diffusion_to_diffusers.py --checkpoint_path model_download/ghostmixV20Fp16.FmfH.safetensors --original_config_file model_download/v1-inference.yaml --image_size 512 --prediction_type epsilon --from_safetensors --dump_path model_download/ghostmix
RuntimeError: Error(s) in loading state_dict for AutoencoderKL:
Missing key(s) in state_dict: "encoder.mid_block.attentions.0.to_q.weight", "encoder.mid_block.attentions.0.to_q.bias", "encoder.mid_block.attentions.0.to_k.weight", "encoder.mid_block.attentions.0.to_k.bias", "encoder.mid_block.attentions.0.to_v.weight", "encoder.mid_block.attentions.0.to_v.bias", "encoder.mid_block.attentions.0.to_out.0.weight", "encoder.mid_block.attentions.0.to_out.0.bias", "decoder.mid_block.attentions.0.to_q.weight", "decoder.mid_block.attentions.0.to_q.bias", "decoder.mid_block.attentions.0.to_k.weight", "decoder.mid_block.attentions.0.to_k.bias", "decoder.mid_block.attentions.0.to_v.weight", "decoder.mid_block.attentions.0.to_v.bias", "decoder.mid_block.attentions.0.to_out.0.weight", "decoder.mid_block.attentions.0.to_out.0.bias".
Unexpected key(s) in state_dict: "encoder.mid_block.attentions.0.key.bias", "encoder.mid_block.attentions.0.key.weight", "encoder.mid_block.attentions.0.proj_attn.bias", "encoder.mid_block.attentions.0.proj_attn.weight", "encoder.mid_block.attentions.0.query.bias", "encoder.mid_block.attentions.0.query.weight", "encoder.mid_block.attentions.0.value.bias", "encoder.mid_block.attentions.0.value.weight", "decoder.mid_block.attentions.0.key.bias", "decoder.mid_block.attentions.0.key.weight", "decoder.mid_block.attentions.0.proj_attn.bias", "decoder.mid_block.attentions.0.proj_attn.weight", "decoder.mid_block.attentions.0.query.bias", "decoder.mid_block.attentions.0.query.weight", "decoder.mid_block.attentions.0.value.bias", "decoder.mid_block.attentions.0.value.weight".
Sorry , I don't know what is it....
知乎来的。哈哈哈哈
哈哈哈,欢迎欢迎~
+1嘿嘿嘿
Dear @Ghost_Shell.
First of all, I really like your works.
Regarding the image reproduction issue, could you please tell me which version you use for "bad_prompt", "bad-artist-anime", "bad-artist" textual inversion?
Because after removing those prompts, I was almost able to reproduce the images.
Thank you in advance.
Thanks for your love. I just feel very sorry about that. Cuz I didn't use any textual inversion relative to "bad_prompt", "bad-artist-anime", "bad-artist".....I copied the negative prompts from others....And everytime I forget to delect that....So if you want to replicate my images , plz just remove the relative textual inversion...you will get the right result.
Hi this is a side question, how long does it take you to generate an image with your 3060ti and let's say 25 steps? Sorry I am just trying to compare performance here. thanks
For compare , there are a lot of things you need to control. So why not just check it on this website: https://docs.getgrist.com/3mjouqRSdkBY/sdperformance It contains almost every performance test.
Gen time on a 3060 (non ti version - 12gb ram) : 768x768 DPM++ 2M Karras @ 25 steps + Upscale to 1536x1536 R-ESRGASN 4x+ Anime6B + Adetailer (for face detail) 1 pass = 1m 52.81s to generate+upscale+adetail.
What does "NoVAE" mean on v2?
It means this version of GhostMix dosen't contain the VAE, you should prepare the VAE by yourself
@Ghost_Shell thank you
大家好，我是作者，如果有任何问题，欢迎来这里留言，在自己能力范围内会尽力给各位解答，并尽力在下个版本帮大家解决。如果大家觉得模型效果不错，请帮忙给模型五星好评，这对我有大的激励作用，感谢。欢迎大家上传自己做的图片，测试模型需要各种各样的Promts，分享Promts可以帮助我做出更好的模型，再次感谢。如果你不会用这个模型，请看模型简介部分。
Hi,everyone.I am the creator of this model, if you have any question or suggestion please tell me here. I will try my best to improve it in the next version.If you like my model, please give me 5 stars review , it encourage me a lot. And you can upload the pics you made,it will help me for testing the next version of model. If you don't know how to use GhostMix, you should check the Instruction part.
i have noticed that your model is realy good at making faces, but that also means i have trouble creating faceless/headless or just someone wearing a helmet/mask. any tips?
@Ponynator For the helme/mask requirement , I think you can check prompts of the 17th & 18th image in my showcase.And as for faceless/headless, I think it is better to find a lora to do so.
Hello,
i have 290k images of faces, can you collaborate, so that we can make it ultra-realistic?
Hi, I love your model. How do you get your images to have that painting and non-realistic quality without the use of lora? There are many prompts posted here where nothing suggests that beautiful art style so I am confused on how you and others got that. I try to put artist names or styles in my prompts but I always get the same type of realistic images instead of stylized and artistic.
@MOONLITBEACH1 Hi,thank you. I think maybe the clip is not right. And the best way is to get to my huggingface page :https://huggingface.co/drnighthan/GhostMix/tree/main download the sample image zip file, unzip it, and directly put the image into the PNG info part of the SD webui to check the prompts
@crnclashers868 actually, GhostMix is not a 3D model , which means it is not good at generating ultra-realistic image. So maybe it is better to find other model creators that create 3D realistic model like BRA,Reliberate,Realistic Vision. I think their model is much better than GhostMix at ultra-realistic area
@Ghost_Shell Thank you
Hi, after I downloaded ghostmix_v20Bakedvae.safetensors and ghostmix_v20Novae.safetensors from the links on the page, I found two files are exactly the same, more specifically ghostmix_v20Novae.safetensors does not have the Sha256 hash as shown on your page, please have a look, thanks.
@xia2willy145 Yes,that's my fault,so sorry....few days ago my fan told me that the No VAE version is the same as the Baked one, I changed the file already. So please download again, I am so sorry...
@_GhostInShell_ If we can work on another project which can specialize generating ultra-realistic models, that would be great
Please, Please don't stop making updates for GhostMix. This is one of my favorite models.
@limblessgirl4172 Thank you for your love. I will continue making the updates for GhostMix. It is quite hard right now. But I will try my best.
Dear GhostinShell, I would like to congratulate you and thank you very much for this very beautiful model. I am very happy that you shared it with us. But I would like to ask you please, why do you have so many limitations to the license? Specifically why can't we sell the images that we generate with your model? I think that stable diffusion is an open sources project? I would like to make a game and I would like to use your model for the images of the game, can I still use it ? Thank you again for sharing your great work.
@_GhostInShell_ Thank you very much for your quick reply! I totally understand. Keep up the great work!
虽然不能给你换4070ti，但是我4090可以给你远程当云主机用
感谢感谢，我618会自己买一块4070，应该不用了
It's written 'NG : Use the model without crediting the creator.' Can you specify when exactly the credit should be posted? Do I need to credit every time I publish an illustration made with ghostmix?
I don't think you need to credit every time , just credit ghostmix the first time you post is fine
Its color looks really grey when I using it? I tried both baked-vae version and no-vae version (with ClearVae and kl-f8-anime vae). What problem could there be?
佬，我用你的模型出的图都看起来灰乎乎的，不知道为啥，不是没加载vae，自带vae和无vae的版本都试了，这可能会是哪里出问题了？😭😭😭
@bluestarr 是2.0吗？试试用我头图的prompts，如果还是灰灰的，那重新下一下BakedVae版本吧，2.0还暂时没有人反应颜色灰这个事
@Ghost_Shell 排查出原因了佬，是Dynamic Thresholding这个插件导致的
@bluestarr ok
这么好的模型竟然是3060ti搞出来的，太厉害了
谢谢好评
Seriously quality here
Thank you
this model is spectacular. i face trained this model and the output is as excellent as the original with the sample prompts provided. bravo.
Thank you
My personal favorite model
Thank you!
我并不确定是不是我的问题，在尝试使用各种对肢体的修复或者组合负面词和文本倒装，还有AFTER DETAILER,这个模型很容易出现一些难以名状掉SAN值的肢体，而且情况相比我同一天尝试跑的其他几个2.5D模型更严重，其他可能只是扭曲或错位，这个会经常性产生一些，我只能称之为克苏鲁严重的四肢情况。。。
是2.0版吗？然后先用样图扔到png info里面看看能不能生成原图先....之前都没有人反应过四肢扭曲特别严重的情况，先把所有修复关了，就开一个高清修复试试
@_GhostInShell_ SiliconThaumaturgy这位油管主测试了模型对手的修复性,似乎他觉得您2.0版本的手产生了负优化的问题.....
@freedomkou 视频我简单看了一下，badhand的negative embedding本身就真不行的，如果感兴趣，可以去nga等论坛查查，我的说明里面也着重说明了这点。他的测试里面没有用easynegative和deepnegative，实际应用当中我推荐这两个会好很多。然后你看Ghostmix基本平面的和2.5D，3D的都能做，而不同风格的手也不一样...所以手差从直观上可以理解。每个人都有自己的观点，我希望做的是一个大兼容的模型，直接出特定风格的好图不是checkpoint应该做的活，而是应该是checkpoint+LoRA+embeddings做的，因为实际应用当中，除非找茬，要不然可能不用像easynegative和deepnegative之类的embeddings
Poses of most nowadays checkpoints are really stiff as if the only purpose of existence is for a portrait.
This model brings back the good old days. (like 8 months ago. 🤤)
Action-packed poses but hard to use, covering various camera angles but hard to use, lots of facial expressions but hard to use, and you could keep discovering new things with this.
a lot of fun.
细节十足，画古典风格的装饰花纹还是这个模型最棒了。不过使用这个模型画场景+人物的时候，很容易出现其他部分都很完美只有人物脸部严重扭曲或者模糊的问题（开了高清修复），请问怎样解决？
有开人脸修复吗？不要开人脸修复，只开高清修复。人脸修复纯粹负优化
用AD修复面部试试
trying to create an armored woman, but the model keeps drawing it in skimpy armor, with thigs exposed, cleavage.... anyone knows any prompt or Lora to deal with this?
try to add the words you don't want the ghostmix create in negative prompt
Bit late to the party, but I believe there's a LoRA on this site called something like "Cold Shower" which should help deal with that - it tries to de-skimpify clothing/armor/etc.
好多说明没看懂
Can you make a real model?
What do you mean ?
What do you mean ?
Okay, I am a little confused and need some help. I downloaded the v2.0.Baked version, but I am getting pretty mediocre results. The eyes and fingers come out very deranged.
I am using the "easynegative" embedding and the highres.fix: 4xUltrasharp_4xUltrasharpV10.
Is that the problem?
I am using auto1111 btw.
What are the prompts and negative prompts? cfg = 7?
@_GhostInShell_
For negative I used "(easynegative:0.6), (worst quality, low quality:1.2), watermark, text, multiple views, bad hands, bad fingers".
As for the prompts: "(masterpiece, top quality, best quality, official art, beautiful and aesthetic:1.2), (1girl:1.3), (fractal art:1.3), <lora:hatsunemiku1:1>, smile, blush, vocaloid3"
Some other info:
"Steps: 20, Sampler: DPM++ 2M Karras, Clip skip: 2, cfg =7 "
I prefer using danbooru tagging. I tried changing my prompts in a multitude of ways, even changing the lora with another character, but it didn't seem to be where the problem was located. My theory is that the upscaler might be the problem?
@Cale5Jubilife sorry, civitai didn't announce me your replied. Yes, upscaler might be a problem. Try to use Highres-Fix with "R-ESRGAN 4x+ Anime6B", 2x upscale rate and 0.4-0.5 denoising strength
Instead of "bad hands, bad fingers" in the negs, use instead (deformed fingers:1.5) (deformed hands:1.5) (amputee:1.5)
话说大哥clip挺好, 挺准确
从海艺过来的，感谢老板做出这么棒的模型。有个小问题想问一下，V2.0-BakedVAE和V2.0-NoVAE有什么区别？
BakedVAE就是内置了VAE，不用设置sdwebui里的VAE也可以出很好的颜色。
请问一下楼主 baked-vae是如何训练得到的？除了滤镜风格外，还会对脸部手部细节进行优化吗？
baked-vae是后续通过sdwebui的工具箱加上去的。VAE解决的是颜色问题，不是解决手部细节优化的。
啥时候搞个国风模型啊大佬
我的观点是ckpt主要解决做得到的问题,比如画风,具体到国风的风格应该是lora来解决的.
作者你好！我使用站内GhostMix V1.2生成的图片发送到文生图试图复现，结果却并不一致。我注意到参考图片里GhostMix V1.2的hash是f20c91bd27，我使用GhostMix V1.2生成的图片hash却是d7465e52e1。我在C站看到GhostMix V1.2的各种hash没有f20c91bd27，想问一下是你后续更改了GhostMix V1.2吗？或者说我想知道f20c91bd27真正对应的模型是哪一个？
你好，是这样的，我一开始1.2生成的图是用unpruned版本的，后面因为模型里面有垃圾，就进行了pruned，所以hash值不一样，但是是可以复现的。不能复现的原因可以查查RNG，随机数生成器，要GPU，而不是CPU，然后就是各种embedding打了没有，应该都是可以复现的。
大佬，你有没有考虑训练一个京阿尼画风的大模型？你的模型好好~
额....我的概念一直是大模型应该解决画风兼容性问题，比如3D,2.5D,2D都能出，而具体的画风问题应该交给LoRA去解决，要固定京阿尼画风就ckpt+lora去实现
有打算试制一个吗？大佬
@VF1 额...没有打算....我还是聚焦在大模型上....京阿尼的lora应该c站有吧
I really enjoyed using your model. It's so great.
It gave me the best result while I was comparing it with other similar models.
Thank you for this awesome model!
Thank you for loving GhostMix !
A great model, salute to you
Thank you!
Awesome model, wich sampler do you think works better with it? thanks
Thank you, I personally perfer DPM++ 2M Karras
nice job
Would you consider an XL/Pony version?
the best model i have used
I'm gonna play with it again with old good memories. 🎉
I tried a lot of models, but your model remains the best ever for my very special vision! thank you so much!
I just added this Checkpoint as a recommended resource for my new lora Luce Is Cute :D
In the days of 1.5 SD, this was probably the best anime model.
Nothing could do backgrounds like this one can.
For those with budget GPUs, this is a highly recommended model.
hello, im new to stable diffusion and a friend recommended this model, im trying to make a male character similar to magical purple dude, but i keep getting squishy girls, i copy pasted everything but the result was also a girl, im not using any lora, or vae, is there anything i can do?
mee to you should to write 1man in positive prompt and negative prompt you should use 2man 2 woman 1 woman vs&