Preview images don't use any other LoRAs.

❌💲 Commercial use of this model is not allowed

This model is new, description is a work in progress.

This model is part off-shoot from Worms Team17 Mayhem. I have rock bottom expectations that anyone besides me would post results using this model, but I went ahead with it anyway because 1) I wanted to see how it would behave with changes to dataset captions (compared to the Mayhem model), and 2) that other model couldn't get his details right (the hair was off and his signature glasses were suppressed by the geek glasses despite different captions).

Sadly, from some tests I ran I already found it suffers the same problems as the Mayhem model, particularly with eyebrows and eyes. AI may still try to give him human ears. Also, the easily noticeable polygons on his moustache lead to weird effects.

🖼DATASET

60 images total, all depicted assets are property of Team17. I reused some pictures from the Mayhem model's dataset but mostly took new ones from W4M gameplay and FMVs (that's also how I noticed his moustache placement in FMVs is different from the actual game lol).

In most images he has brown eyes and is wearing glasses. The glasses' lens vary between white (in game) and transparent (FMV). The few WUM pictures have grey eyes.

How is captioning here different from the Mayhem model?

🔄REPLACED:

• "half-closed eyes" --> "lidded eyes"

• "black rounded eyebrows" --> "floating black rounded eyebrows"

• "turned away from viewer" --> "facing away"

• "belly up" --> "on back"

• captions that said "fist" now say "clenched hand"

➖REMOVED:

• head angle captions (some time after training the Mayhem model I began thinking they're useless + worms' faces tend to get deformed at complex angles anyway)

• "looking at viewer" as an experiment

• some other captions for the sake of simplicity / due to redundance

➕ADDED:

• "elderly male" (I didn't use gender and age captions for the Mayhem model's dataset)

• "small pupils" as an experiment

• "tail" where it can be easily seen (but it may not be the case with generation results)

• captions for eye colour

• captions for things in background

• "day", "dawn/sunset" and "night" (night may need to be reinforced with additional captions)