stinkywormynoob

[stinkywormynoob]

When installing Worms 3D I saw some pictures of worms but I didn't snap them all and I can't find them in the game's files, so I decided to try reverse image search. And just like many other rev. image searches with obscure images it turned out to be utterly useless. And the very first picture here is my own fanart.

[stinkywormynoob]

[stinkywormynoob]

Pure AI on the left, manual edit on the right.

I only really use Photoshop for art, not editing, I hope my lack of editing experience doesn't show much. I decided to try improving some AI-generated pictures for Civitai, although it's mostly things that are relatively easy to fix. Out of several pictures I touched up, this one changed the most.

[stinkywormynoob]

[stinkywormynoob]

Damn, is there anything I can do with inpainting? This is just one of the multiple attempts. AI keeps the hair short but tends to add human face features.

I also tried using it to change the colour of horns, but instead it removed them and tried to add a devil woman in place of one horn.

[stinkywormynoob]

Trying to use inpainting to remove unwanted elements in pictures (I'm working on a dataset for another Worms model) took an unexpected turn. If I can't convince AI to simply get rid of the other worm and NOT add any other unwanted stuff in his place, I'll have to do a crude Brush work. I checked the Clone Stamp tool today and didn't understand shit. I don't remember the 3rd pic also (I interrupted the other generations)???

Sure, AI, this is what yellow and orange looks like.

And here I thought I could use AI to carefully remove extra pupils (the Inpaint Additional Prompt was "nothing")... I guess it's the good old raster graphics software for me.

[stinkywormynoob]

Trained another Worms model. For now I'm only trying out source accurate generations, but I want to experiment later.

Prompts:

wormsteam17, mayhemworm, worm, 3d, limbless, armless, legless, solo, no humans, 3/4 view face, professor worminkle, black rounded eyebrows, white hair, balding, moustache, half-closed eyes, frown, closed mouth, handless, half moon reading glasses with golden frame, simple street background

wormsteam17, mayhemworm, worm, 3d, limbless, armless, legless, solo, no humans, front view face, default worm, black rounded eyebrows, 2 disembodied hands, looking at viewer, open mouth, upper teeth only, tongue, simple background

wormsteam17, mayhemworm, worm, 3d, limbless, armless, legless, solo, no humans, 3/4 view face, eyebrowless, pink hair, short twintails, bangs, white hairbows, elvis presley sunglasses with white frame and purple lens, 1 disembodied hand with white glove, waving, smile, simple nature background

[stinkywormynoob]

I recently found out about weights.gg and decided to upload my Worms voice models there, it seems more user-friendly. You can't use inference or download models without an account though.

And this night I dreamt they took down my Whoopsie models.

[stinkywormynoob]

updated (sorry about faded colors, there was a big clean up on my PC and it reset my settings)

tiering is hard, IDK

[stinkywormynoob]

[stinkywormynoob]

[stinkywormynoob]

[stinkywormynoob]

[stinkywormynoob]

autotagging was a mistake

I wanted to try using it to add a couple captions to all images in the dataset in the end, but it created a mess which includes double "solo" tags and captioning some images with "pokemon (creature)"

Edit: looks like the autotagging only affected about 1/3 of the dataset (which currently takes 90 pages)

Edit2: nvm there are more aaaaaaaaaaaa

[stinkywormynoob]

Attempts to generate a worm with humanoid body wearing a tracksuit in the night streets (base model DynaVision XL + 2 LoRAs besides my Worms one):

I'm cherrypicking this one:

Other attempts below:

Just a weird human

Head looks semi-human

The worm head is here but missing pupils

Eyes are complete, but the legs are way too short (also suddenly off shoulder o_o)

[stinkywormynoob]

Solo worm generation results (base model: DynaVision XL)

[stinkywormynoob]

Multiple worms generation results (base model: DynaVision XL)