AI Image Generation

Wanted to see if the model mix I'm using (ares / chillout) would know my beloved Abigail from Stardew Valley. Here's what she looks like in game:


The prompt is just Abigail from Stardew Valley:


It almost seems like it knows that Purple is involved, but it didn't make it to the hair. The backgrounds are so good though!

Here is with Realistic and Purple hair added:


Notice how this one added a bunch of lavender. I think this is the purple prompt leaking over into the scenery:


It's ok, but I think I'd probably have more luck just describing her verbosely, but cool that it actually knows what Stardew is.

Hmm adding the word goth anywhere creates completely untamable moon sized mammaries.

not really ai but cool update and i can see great use cases for artists who get really efficient with ai tools just pumping out ridiculous amounts of gorgeous assets for game dev. unity and unreals asset library is about to explode exponentially

Not having much luck. There are so many elements to the outfit that it gets all the colours mixed up.

A mix of ares and chillout models. I mainly mixed them to get a VAE in there. There is a way to load it separately but I haven't figured it out yet.

The above more cartoony is chillout mixed with an anime model.

BTW Ares and Chillout both seem to favour big noses, and when you try to tone it down in the prompt you get young people, so I often have to put young synonyms in the negative prompt.

Tried to upscale an EQ screenshot with Midjourney but results are kinda all over. It seems to recognize the house and character but the rest of the setting gets lost a bit. It often turns into into an indoor scene and it doesn't really recognize the armour as such.








Trying with just text descriptions gets a more photorealistic image but I can't get it to make the character hold a flail, size is always way off or it's just a distorted sword. Guess it's not trained as well as swords are in the data set.

I'll grab your image and try it with stable diffusion's SD Upscale script, I'm curious.

Forgetting about the weapon for a bit and just trying for purple plate



TRAIN!

The upscale was really bad.

Here's img2img with .3 denoising:



.4


.5

.6

I used the prompt: Armored girl standing behind a monument, cottage, rocky forested hills, flaming sword, grassland, clear sky, purple plate armor

Kharzette said:

The upscale was really bad.

Here's img2img with .3 denoising:



I used the prompt: Armored girl standing behind a monument, cottage, rocky forested hills, flaming sword, grassland, clear sky, purple plate armor

Click to expand...

I like the upscale on the environment on the first two. Maybe the character model is too small in the picture. I tried again on midjourney with a cropped picture with a simpler background and it did a lot better, especially if the input included the desired output. Seems to mirror them most of the time for some reason.



Input for this was base image + description "female elf holding black sword"



Lots had strange posture, this might be because the original EQ model has kind of a wierd stance in the first place (massive boobs, head fairly far back)



Same input image with more realistic direction

Very nice! Yea luclin models have such terrible poses and movement, might have more luck with old models.

I let a batch of 100 run and none were any better than the above.

Lambourne ok I challenge you to this I know you can do it, I wanna see a lifelike version of a raid cheesing one of the warders in sleepers tomb with their clerics position

Pasteton said:

Lambourne ok I challenge you to this I know you can do it, I wanna see a lifelike version of a raid cheesing one of the warders in sleepers tomb with their clerics position

Click to expand...

Never actually done any ST raids so not sure what the strat involves or even what the room looks like. It's also hard to get it to position figures in a particular position if there are many. But will try and go off an image from alla:







More lifelike/cinematic style







Had to resize the last two because they are getting too large to attach, quality suffered a little bit.

Making some Mercy cosplayers while I watch the league matches today. Can't quite get a good Mercy face exactly.

Some cosplayer selfies. I wish I could do multiple characters but when I do it mixes them together.

I got bored and threw a big jumble of overwatch related stuff in a prompt. The result is a new fps, Overwatch: DERP SQUAD

Holy shit those are pretty good. Haha id love to see an avatar of war raid with a ridiculous Zerg of people and tons of dea warriors on the ground but doubt the ai will know what avatar of war looks like