santa hat
deerdeer nosedeer glow
Sign In

My AI Generations With Issues

Hi everyone. Just quick information of what my settings & models / vae's are before I dive into this.

I use:
Anything 4.5
AOM (AbbysOrangeMix)

Vae's I'm using are:
Pastel Waifu Diffusion

I mostly use 512 x 856
20 to 40 steps
7 to 10 CFG

Euler A & DPM ++SDE Karras

Lately I've realised that my anime prompts have become somewhat looking '' bad quality '' as in they're not looking HD , low quality , fuzzyness, problems with the eyes, eyes glitching out, skin not looking that good. I've looked into the negative prompts & tried dozens of them, tried different settings, different height & width.

I'm wondering why because I don't think i've had this problem before. I know the eye glitching thingy was always there but now the issue is, it's almost on every model now unless I really focus the prompts on the eyes (close up). Nowadays I have to scale up every image I like or scale them up so they look better.

I see multiple anime prompts on civitai and other platforms and I'm really wondering as to how you guys or I my self can create a nice prompt as in (Quality wise, Skin looks good, Eyes looking good)

I can show you some of my prompts of what I mean. (Contains NSFW)

^ Eyes lookg bad / glitchy, the overal quality is fuzzy.

^ Reduced Quality,Bad Pixels,Bad Eyes

^ This is a Lora I used, the image it selfs look lovely. But the image as a quality looks bad, pixels lowered, the smoothness isn't there, now if I compare this to my latest prompt of the same image bellow.

^ As you may have noticed, the image looks way more clear here, pixels on point. Eyes looking good, so does the skin and overal generation of this image.

So am I just overseeing things or am I doing something wrong ?

What can I do to prompt more beautiful art, what makes it so one image is better than the other ?

I've prompted the same image again and I can clearly see loss of quality.. that makes me sad

I don't know if ya'll can see the difference :c

1 Answer

I've been trying to compare the last two pics you posted and I can't get the same results. I can tell you you don't actually have embeddings you use I think. I tried with and without them and I'm getting same poses as you only without them. Also you say you prompted those pictures the same but actually the second one have higher Lora weight. Which could be the cause of this, but for me it still looks sharp in both cases. I can't compare the VAE used through so your problem might be there. Or it's something else that doesn't show on metadata.

Also just to point out, you used different resolution on both.

Your answer