[DeepCobalt]
Recommended Parameters:
Sampler: Euler a
CFG Scale: 3~4
Steps: 20~28
Clip Skip : 2
[DeepNavy]
Recommended Parameters:
Sampler: DPM++_2M_SDE
CFG Scale: 4
Steps: 24
Clip Skip : 2
[DeepBlue]
Recommended Parameters:
Sampler: Euler a
CFG Scale: 3 – 5
Steps: 20 – 30
Clip Skip : 2
[DeepViolet]
Vpred Model
Description
I worked on removing the noise that used to appear around the eyes and nose in previous versions. As a result, there are quite a few differences from the original DeepBlue model, but I feel that this version understands prompts much better.
You can now use the Euler a sampler without getting any of that noise.
FAQ
Comments (21)
Takes a little time to set up, but produces great results. Very satisfied! 💕
I may not use the Model for the intended purpose, but I love the style and prompt adherence!
I love the model but I am tired of the auction 😫
Hey @Mol_Keun, are you rendering your stuff out in full rez or do you use Hires. fix?
If you use Hires. fix, do you have any recommended settings?
Love the model, these are by far some of the best results i get in this segment!1!!! <3
It's a shame that DeebBlue V2 became unavailable. It feels like a better version of DeepNavy when you use them back to back
I agree. In my case I think DeepBlue V2 is a way prettier and well designed version than the new DeepNavy. I gave it a lot of chances but it's just not as good for me. It's a pity we can't use it anymore. Might just use another model.
I'm sorry that the DeepBlue model can’t be used for the time being. Until now, I had to manually register it every Friday by using Buzz on the auction system for a week of access.
But this time, it looks like Civitai made things a bit more convenient—there’s now a feature that automates that process!
If everything goes smoothly, I should be able to keep using the model consistently starting next week.
Mol_Keun Take your time, don't worry. Huge thanks for notifying us. 😊
Mol_Keun Thank you for the info!
But DeepBlue V2 is DeepBlue EPS but with a different name, right? It's still available for download.
hi @Mol_Keun, please answer me, how do you generate such images? No matter how much I tried, I still can't generate such high-quality pictures. Your pictures look alive, have small details, different emotions that suit the situation, it looks like ai know what's in the picture and recognizes it. Please share your settings or where you generate, both are possible. I need at least some example by which I can make them :(
I am very grateful to you for such a wonderful model.
I am not Mol Keun but I would recommend that you copy prompts from good quality images. For example the images that Mol Keun posted have the prompt in the image code and can be seen if you put those images in PNG info in Automatic1111.
But if you want to use only CivitAI you can look at the images I have posted or other people's and copy the prompts and see if you have favorable results. Finally I would recommend that you use the parameters, quality prompts and negative prompts that Mol Keun recommends in the description of each model, and above all that you should use Hi-res Fix and upscale to fix any artifacts that are generated (most likely artifacts will continue to be generated with Hi-res, so you will have to repeat the process several times as it is trial and error).
Kenmuri I've heard of automatic1111, of course, but I've never tried it, I use comfyui. I'll try to put an image into automatic1111 sometime, maybe I'll see something.
thanks for the advice
If you are using Comfy UI, there is no need to switch to A1111. If you want more dynamic poses, specify the pose in the prompt or try using ControlNet or I2I.
Details can be enhanced through Hires, or you can crop only a local part of the image, add a bit of noise with Ksampler, and refine it.
Well, I'm mostly having trouble getting information. Do you know any sources I can get it from? Anything will do. Preferably a description of the nodes, what each parameter does, etc. Guide or something like that
Kenko_gras A1111 and Comfyui will generate different results even with the same settings. This is because the tokens (words in the prompt) are weighted differently between the two programs. I believe Civitai's gen uses A1111's token wieghts. But there was a time they tried out comfyui's wieghts and people disliked it because of how differently the results were.
I highly recommend trying out A1111 for a while before deciding to switch to comfyui.
Comfyui can technically do better with upscaling and giving hi-res images. Its also faster too
But A1111 gives much more creative and aesthetic results that seem more expressive.
It's good at basic use but bad at animal genitalia and non-navel. It's highly biased.
Does anyone know how to get cetacean penis and remove navel ?
I don't know about the navel other than adding it to negative prompts. Also, this will help with the cetacean penis: https://civitai.com/models/1173463?modelVersionId=1329472
@PurpleCrystalFox
Thank you for your response! I'll try it.
It seems that the navel comes from the word concerns "belly" or something, so when I prevent to use those words it works good. (using "plump" instead of "big belly" etc.)
using the from_behind tag seems to add some sort of rod or figure clamp going through the neck or the spine of the character from time to time on both bluev2 and navy. is this bleeding from some lora in the merge?
It's really good, it just needs some improvements with text comprehension, but it's really good.