❌? Commercial use of this model is not allowed

Preview images use different checkpoints, but no other LoRAs.

I apologise for the Fooocus tryout images (especially screenshots) in the preview, but I cannot remove them in a convenient way.

⚠WARNINGS

• DO NOT USE WITH PONY CHECKPOINTS OR YOU'LL GET STUFF LIKE THIS. I've trained a few other Worms LoRAs for Pony.

• Common issues: bad or missing mouths, double tails (especially if the worm's "chest" is facing the viewer), wrong number of eyebrows or pupils (or missing entirely); less often but may occur randomly - conjoined worms.

• The worm's face may get distorted and get human features when combined with models trained on human faces, or if you set this model's weight too low, or with certain prompts. He may also gain complete arms as opposed to the usual floating hands.

• Controlling the worm's eyelids is near impossible, at random rare occasions the worm's eyes are not fully open.

• Heavy worms turn out the least accurate. I tried to caption their bodies and lower lips but I don't know how to caption their head shape.

• Asking for multiple worms may result in ambiguous mess under/behind worms. The AI probably won't make the right number of worms if you ask for any specific number above one. Multiple worms are also generally less coherent.

• In-universe items will most likely look bad, but this is meant to be a character model anyway.

ℹ USE HINTS

• Dataset includes unusually shaped worms (scientists and heavies). To avoid massive craniums of scientists, put "scientist class worm, big cranium" into negative prompt. To avoid large bodies of heavies, put "heavy class worm, big body" or something similar into negative prompt.

• You can do a bit of in-universe style manipulation with "worms 3d render", "worms forts under siege render" and "worms g4 render", but the worm's body is usually wrinkly.

• There's a little bit of text in the dataset so you might want to exclude "text".

• I've made a list of checkpoints which I think make cartoony 3D style pictures well, but anything 3D specialised should be good. I don't recommend realism checkpoints as their results might be slightly uncanny.

?DATASET

Trained on 53 images of Worms renders (all captioned), mainly those related to:

• Worms 3D

• Worms Forts: Under Siege

• Worms: Revolution /Clan Wars /Battlegrounds (G4 games where worms come in different shapes)

... so the way worms look in the dataset images have some variety. Some pictures have 2 or more worms which were confusing to tag, maybe I shouldn't have even included them ?‍♀️

❌ Dataset does NOT include:

• G1 material

• G2 material

• Worms Rumble material

• any in-game images

• any fanmade models (all images are property of Team17, with VERY minor edits)

描述:

Trained on SDXL.

This is actually a 2nd training attempt. The DDIM sampler made the epoch test results look like garbage and I wasted Buzz thinking the 1st training attempt was a failure because of it. >:( DO ? NOT ? TRUST ? THE ? SAMPLES ?

训练词语: worm, render