Midnight's AI Image Peddler Booth

[Midnight-Phantom]

Can you make a picture of a high elf woman sitting on a throne?

The place must look like a fantastical ancient ruin with advanced elven magic.

She's wearing a long dress. Still not sure about her hair or eye color.

yep ..

Spoiler: ....

 

[RepresentingEnvy]

Can I get an artist teddy bear wearing shades? For no reason at all.

Edit: And dollar signs in the image.

 

[Midnight-Phantom]

what about the cute anthropomorphic rat girl with white fur and red eyes?

mmmm ...

Spoiler: ....

 

[HokuouTenrou]

Hmm, interesting. You did mention that you use 'additional techniques', so like photoshop or something? Kinda curious with your workflow.

 

[ACertainPassingUser]

Can you make group of women in the beach, but they're using modest full body female swimming suit instead of bikini ?

I'm rather bored with usual "anime beach episode" where they're using "un-modest" bikini for fanservice. I always use full body swimming suit to increase my mobility and protect my skin in the pool and especially beach.

I would like to see full enthusiastic female divers using full body swimming suit to protect their body during serious swimming.

And also, can you make the same thing above, but in the pool instead of beach ?

Image references :

Spoiler

 

[beast_regards]

mmmm ...

Not quite I had in mind, but it went far much much better than I expected.

 

[Midnight-Phantom]

Hmm, interesting. You did mention that you use 'additional techniques', so like photoshop or something? Kinda curious with your workflow.

Naah naahh bro, these days, I'm into using GIMP and still learning the ropes. In the past, I used Photoshop for some light editing and stuff but they become too much gready and launched subscriptions. Nowadays, I do use GIMP occasionally, mostly for control net and stuff, but I don't spend too much time on it if I don't feel like investing, you know?, if you're willing to put in the work and try out different platforms, you can create all sorts of images.

For generating images, my go-to is usually automatic 1111 or comfy UI - they do a pretty good job. When it comes to editing, I rely on GIMP, and sometimes I use Krita for hands and stuff; it's pretty user-friendly, But to be honest, I don't use them unless I want to put some serious effort into a specific image.

Recently, I've been learning a lot since I started generating images on my own device, about a month or so ago. There's still a ton of stuff I need to check out. Usually, I just cycle through images with different LORA and settings, upscale them, and post the ones with cool overall visuals. Different prompts require a slightly different approach, and even tweaking variables like VAEs, clip skip, and the amount of LORA can have a big impact, though it could also be due to my inexperience.

My device is pretty powerful, not bleeding edge but definitely up there in the spectrum - it's an year and half old build using an RTX 3070 and likewise, some pretty good other new-gen components, which gives me a smooth experience overall. But I've seen folks on YouTube rocking those 4090s and other powerful setups, and they churn out images super fast with more complex workflows.

I try to get the most out of my device, but running 30-40 upscales in one go would take forever and put too much strain on it. So, in those cases, I adjust and do lighter batches, and then move on to other stuff. Gotta keep things running smoothly, without straining my babe too much you know ..

Not quite I had in mind, but it went far much much better than I expected.

I thought about doing full furry images, too, with minimal human features. But I don't know why I like the ones with more human features and some furry features, so I posted them with a few tweaks here and there.

To be honest, generating images with AI is fun and fast, but really creating a detailed one takes a couple of days of work on a single image to make it good. It takes time to get those perfect images, or by luck you may get one just based on the models or additional lora and stuff you put it on.

I plan to do the same prompts again as you and other people listed when I have free time. I will probably post some too, and if I do, I will keep in mind to tag you or others here so you can check them out then.

I am still inexperienced in generative AI art, and to be honest, I mostly tweaked the images that I personally favoured. So I probably can't live up to expectations, but thanks for giving me ideas for generating images. You know, it's hard for me to come up with characters.

 

[beast_regards]

I thought about doing full furry images, too, with minimal human features. But I don't know why I like the ones with more human features and some furry features, so I posted them with a few tweaks here and there.

Kemonomimi is just far more common for anime. AI struggles with anthros, it's much more difficult, but occasionally it manages. I am not sure which one you use, I've seen many managed to do it even with the AI.

 

[HokuouTenrou]

For generating images, my go-to is usually automatic 1111 or comfy UI - they do a pretty good job. When it comes to editing, I rely on GIMP, and sometimes I use Krita for hands and stuff; it's pretty user-friendly, But to be honest, I don't use them unless I want to put some serious effort into a specific image.

Ah, gotcha, so you did some touch up. Actually interested to request some, but I dunno if images that I generated on my own (for reference) are acceptable here, instead of text prompts like the other members. (I use NoCrypt, for free...)

 

[Midnight-Phantom]

Kemonomimi is just far more common for anime. AI struggles with anthros, it's much more difficult, but occasionally it manages. I am not sure which one you use, I've seen many managed to do it even with the AI.

Nowadays, there are a lot of furry models and LORAs with furry characters. You should check them out in Civit ai. These models are specifically designed for furry types, but they cannot create normal characters well

Ah, gotcha, so you did some touch up. Actually interested to request some, but I dunno if images that I generated on my own (for reference) are acceptable here, instead of text prompts like the other members. (I use NoCrypt, for free...)

Sure you can ..

 

[HokuouTenrou]

Sure you can ..

Oh, nice, I don't know much I can request, but here goes.

Spoiler: ref image

Spoiler: prompts and other stuff

<lora:komori-style_v3.0:1>, best quality, masterpiece, 1girl, solo, solo focus, \(very long hair\), blonde hair, \(drill locks\), blue eyes, large breasts, cleavage, upper body, curvy, mature female, pink lips, \(bridal tiara\), \(strapless ballgown wedding dress\), \(white elbow gloves\), looking at viewer, facing viewer, standing, elegant, relaxed

Model used: abyss orange mix latest ver? civitai 17233
LORA used: Komori Kei art style; civitai 103631

my problem is whenever is use high-res fix or img2img, inevitably there's going to be some oddities, like "man, the hair shouldn't go like this", or "what is that piece of fabric doing there?", something like that.
the positive prompt is probably different from the first time I generated.

 

[Midnight-Phantom]

Can I get an artist teddy bear wearing shades? For no reason at all.

Edit: And dollar signs in the image.

here you go envy ..

Spoiler: ....]

Oh, nice, I don't know much I can request, but here goes.

Spoiler: ref image

View attachment 20713

Spoiler: prompts and other stuff

<lora:komori-style_v3.0:1>, best quality, masterpiece, 1girl, solo, solo focus, \(very long hair\), blonde hair, \(drill locks\), blue eyes, large breasts, cleavage, upper body, curvy, mature female, pink lips, \(bridal tiara\), \(strapless ballgown wedding dress\), \(white elbow gloves\), looking at viewer, facing viewer, standing, elegant, relaxed

Model used: abyss orange mix latest ver? civitai 17233
LORA used: Komori Kei art style; civitai 103631

my problem is whenever is use high-res fix or img2img, inevitably there's going to be some oddities, like "man, the hair shouldn't go like this", or "what is that piece of fabric doing there?", something like that.
the positive prompt is probably different from the first time I generated.

What do you want to do with the image tho .. ?

 

[RepresentingEnvy]

here you go envy ..

Spoiler: ....]

What do you want to do with the image tho .. ?

I just wanted to portray the resident forum teddy bruh. @Hans.Trondheim

 

[Midnight-Phantom]

I just wanted to portray the resident forum teddy bruh. @Hans.Trondheim

oh ho .. so that's why ..

 

[HokuouTenrou]

What do you want to do with the image tho .. ?

Oh, sorry, should've specified in the first place. So I want a picture of 'beautiful blonde girl in a wedding dress'.
Using the reference, I'd like you to either hi-res fix or img2img it, and retouch it if some oddities appear; sudden bad hands out of nowhere, wonky eyebrows, bad hair etc.

 

[Midnight-Phantom]

Oh, nice, I don't know much I can request, but here goes.

Spoiler: ref image

View attachment 20713

Spoiler: prompts and other stuff

<lora:komori-style_v3.0:1>, best quality, masterpiece, 1girl, solo, solo focus, \(very long hair\), blonde hair, \(drill locks\), blue eyes, large breasts, cleavage, upper body, curvy, mature female, pink lips, \(bridal tiara\), \(strapless ballgown wedding dress\), \(white elbow gloves\), looking at viewer, facing viewer, standing, elegant, relaxed

Model used: abyss orange mix latest ver? civitai 17233
LORA used: Komori Kei art style; civitai 103631

my problem is whenever is use high-res fix or img2img, inevitably there's going to be some oddities, like "man, the hair shouldn't go like this", or "what is that piece of fabric doing there?", something like that.
the positive prompt is probably different from the first time I generated.

Here you go bro. I used the same model, lora, and prompts from the attached image and the one you mentioned, for the following generations. I only added some extra stuff, and then I upscaled them.

Spoiler: <->

This one is without the artist Lora, turned out quite good tho ..

Spoiler: <->

Oh, sorry, should've specified in the first place. So I want a picture of 'beautiful blonde girl in a wedding dress'.
Using the reference, I'd like you to either hi-res fix or img2img it, and retouch it if some oddities appear; sudden bad hands out of nowhere, wonky eyebrows, bad hair etc.

.. check them out ..

 

[HokuouTenrou]

Here you go bro. I used the same model, lora, and prompts from the attached image and the one you mentioned, for the following generations. I only added some extra stuff, and then I upscaled them.

Thanks for the work! I like em big.
Ahem, but yeah, the artist LORA is necessary; I just love the art style.
By the way, you don't use any VAE? The colors look faded.

 

[greyblob]

Oh, sorry, should've specified in the first place. So I want a picture of 'beautiful blonde girl in a wedding dress'.
Using the reference, I'd like you to either hi-res fix or img2img it, and retouch it if some oddities appear; sudden bad hands out of nowhere, wonky eyebrows, bad hair etc.

if you do img2img at very low denoising the image remains relatively the same.

 

[HokuouTenrou]

if you do img2img at very low denoising the image remains relatively the same.

What number is recommended? I use either 5.5 or 6; lower than that and there's increased risk of... artifacts?

 

[greyblob]

What number is recommended? I use either 5.5 or 6; lower than that and there's increased risk of... artifacts?

.1 ~ .2 if you want almost no changes. .3 ~ .4 if you want some small differences

edit: this is for upscaling correct?