It looks like a great model,Can I download this model from other websites or channels?

@hakubun have you downloaded it yet? If not, I will upload it to GG Drive and you can download it through there

use vae-ft-mse-840000-ema-pruned.safetensors

What's the prompt?

@Nekos4Lyfe I am new about stable diffusion. So I am sorry if the prompt is not good and also the negative prompt.So I'm doing trials and tests and so I'm always changing and modifying. I'm going crazy.

The pompt is:

model 3d cat full body, white cat, Pearl eyes high realistic blue, body fused and mixer with leafes color gold, neon blue, green and purple and red flowers in ears in middle of arms and in the tail. The tail is with gold augumentation and attachment beautiful and filament. Black background, realistic, smooth, (realistic fur), 3d art, 3d, intricate details , octane render + unreal engine 5 - w 1024 n, (vibrant color), elegant, ((soft light))

The negative prompt is:

nsfw, cartoon, lowres, bad hands, (text, artist name), error, missing fingers, fewer digits, cropped, worst quality, low quality. ((mutated, deformed, extra head, extra arm, extra leg, extra digit, extra nose, extra mounth, extra eyes, extra tail, extra paws, extra feet, bad anatomy, extra finger)), cloned face, dehydrated, duplicate, owres, poorly drawn face, picture frame, framework, disfigured, (( semi-realistic, cgi, sketch, drawing, painting, blurry, blurred)) humanoid, human, brand, brand mark, brand name, ((username)), ((signature)), (worst quality:2), (low quality:2), (normal quality:2), worst quality, low quality, jpeg artifacts, ugly, [deformed iris, deformed pupils], long body

I have problem for extra body parts. texts, light too bright.

I'm going crazy. If you want u can find me on Discord with : NeosBlack#8832

@NeosBlack Hey, Instead of using extremely long negative prompt, try some negative embeddings. Those usually help with weird stuff. Another thing i have noticed is, it is not good idea to use anything with strength of (x:2). Can cause some weird stuff..

@potatovision842 Good advice! I haven't tried negative embeddings myself. Might be why some of my art so far is a bit gray lol

Me and NeosBlack did revise the prompt mentioned above. Feel free to leave suggestions.

Using 30 steps for Euler a :

prompt: " [blurry black pattern : white tail : 3]

[ photo white cat body cat photo Pearl eyes blue leaves gold leaf neon blue green purple flowers red ears gold augmentation : studio 8k ((hyperrealism)) : 15 ]

[billowing : . : 10]

[ : fur : 20 ]

negative

" ((cat-tail)) nsfw

[

:

nsfw, cartoon, lowres, bad hands, text, artist name, error, missing fingers, fewer digits, cropped, worst quality, low quality. mutated, deformed, extra head, extra arm, extra leg, extra digit, extra nose, extra mounth, extra eyes, extra tail, extra paws, extra feet, bad anatomy, extra finger, cloned face, dehydrated, duplicate, owres, poorly drawn face, picture frame, framework, disfigured, semi-realistic, cgi, sketch, drawing, painting, blurry, blurred humanoid, human, brand, brand mark, brand name,, worst quality, low quality, jpeg artifacts, ugly, deformed iris, deformed pupils , long body

: 10

] "

@Nekos4Lyfe Hey, i'll try to generate this and get back to you.

@NeosBlack @Nekos4Lyfe Hey,

Found little time to figure this out now.

I noticed 2 problems that might be causing it.

1. Clip skip. You can change clip skip from settings -> stable diffusion -> clip skip. Usually clip skip 2 works best for most of the models. It depends on model.

You can get clip skip to the top of ui for fast change. Having VAE there is also good idea. This can be done from settings -> user interface -> Quicksettings list, and adding this to there:

sd_model_checkpoint, CLIP_stop_at_last_layers,sd_vae

you need to reload after this

It looks like this: https://i.imgur.com/L5WxkJ2.png

2. Having cat 2 times in the prompt

It is not good idea to have duplicats of words in the prompt. It can be causing the cat to try to generate twice and multiplying it's bodyparts.

Here is fast test i did for the clip skip and prompt neos gave: https://i.imgur.com/VixoJLQ.png . As the tail is still twice in the prompt, did not notice that. It generated 2 tails :) (this is book case example of why you should not copy paste prompts ;))

For negative i had:

(low quality, worst quality:1.4), verybadimagenegative_v1.3, (Unspeakable-Horrors-Composition-4v:1.0),

Both of these embeddings can be found from civitai.

Tips for prompting:

The arrangement of prompt matters, words first will have higher impact on the image than last.

Don't make prompt overly complicated.

Learn to build prompt from empty prompt and understand what everything does to the image, generating between adding large parts of prompts is good idea.

Using a lot of different colors is not going to work, it is going to mess the colors around in parts you don't want them in. Using extension called https://github.com/hnmr293/sd-webui-cutoff can help with this somewhat.

Good luck, let me know if there is still something bothering.

@potatovision842 I didn't have a moment to read everything and make some. Proofs. As soon as I can ( as soon as possible, I will get on it and inform you, thank you very much)

@potatovision842 
651 / 3000

Traduzione

Okay I read everything and thank you very much for the help. I need to improve the prompt as I have already been told and I think I understand. In any case I read that a VAE might help. Could you recommend something that might suit me regarding this kind of fantasy animal with modified body or fused with other objects? And more. For the embeds that you put in, do I just look for that name written in your negative prompt? I need to see about editing the clip skip as well. You can tell I am new to this ahaha. One thing I don't really like is the poses that come out for the most part, not all of them be clear. But many of them are not the best. tomorrow i'll try make this change

@NeosBlack You welcome. Personally i like https://civitai.com/models/22354/clearvae . I think the most popular one is vae-ft-mse-840000-ema-pruned and can be found from stability ai huggingface page.

These are the negatives embeddings i use:

https://civitai.com/models/11772/verybadimagenegative

https://civitai.com/models/4499/unspeakable-horrors-negative-prompt

You can find more with just typing "negative" at the search bar and see if there is something better for you.

There is extension called open pose editor for posing if you want to get better poses heh https://github.com/nonnonstop/sd-webui-3d-open-pose-editor

The clip skip is really important don't skip that ;)

@potatovision842 apart from the fail above which shows my use of the translator hahaha. okay thank you. to install the VAE if I'm not mistaken you wrote about how to do it right? as soon as I have a moment of time I'll do everything and let you know if there's any news. if you want in case write me on discord ( NeosBlack # 8832 ) so in case I show you the final results

@potatovision842 one question. For negative horror.which is better? I see only 4 vector trained, 16... 32.. etc ( and only composition) there is only 1 download or i need download only one this? ( Sorry i'm with phone for now)

@NeosBlack I use the 4 vectors one, it adds less negative and seems to work pretty well without changing the original image too much. Too much, too strong negatives want to derail your image from your prompt. (This is personal opinion from testing)

VAE has to be dropped to the models/VAE folder. After that you can select it in the settings -> stable diffusion. I recommend adding it to the top of the bar for quick switching like i mentioned in earlier comment.

Hope these comments helped you :)

@NeosBlack I use the 4 vectors one, it adds less negative and seems to work pretty well without changing the original image too much. Too much, too strong negatives want to derail your image from your prompt. (This is personal opinion from testing)

VAE has to be dropped to the models/VAE folder. After that you can select it in the settings -> stable diffusion. I recommend adding it to the top of the bar for quick switching like i mentioned in earlier comment.

Hope these comments helped you :)

@potatovision842  yes, thank you very much <3 really. you are really great people. thank you again. I just have to be able to find Cutoff, because I can't find it in the extensions

@NeosBlack The cutoff is extension and is not necessary, i just mentioned it, as it is good for use of many different colors. Seemed like you were interested in having a lot of different colors looking a the prompt. Might only work for clothes though. Not 100% sure. Tested it a while back and seemed to work.

@potatovision842 I have now had time to do some tests. they are wonderful, much better than before. but it still always has some problems. too many tails (even though I removed the double input in the prompt) and occasionally some extra paws. it also happened that elongated and enormously tall cats came. (i modified everything as you said). don't know what to think honestly

@NeosBlack Have you tried with different models? some excels better at making different things than other. Also different samplers give different results, some samplers need more steps and some less.

@NeosBlack Also some quality tags could be causing some randomness.

@potatovision842 Yes, I have tried several models and so far this is the one I like the most

Sorry to bother again in this very long comment. Would anyone have any embbending to exclude furries?

@NeosBlack I think you have to take those out with tags, embeddings mainly fix lines and stuff.

@potatovision842 okay but i tried, on the negative prompt: human, humanoid, furries, furry. But nothing

@NeosBlack Not really expert on furry wording, one word is anthro you could negative. Maybe you have something in positive that leans more into furry than proper cat. For example some cat parts like "paws" could be causing this etc.. It depends a lot on model where it leans on what words..

Hope you getting better stuff from this helping :P

He/she using Deliberate mixed...i assume the same vae.

use vae-ft-mse-840000-ema-pruned.safetensors