This is a test to see if reducing the dimension to 8 is enough to use non-ideal images. If this works, then i might not need to use high qualty anime style images to get an anime output, just drawings in general.Trained on 24 epochs, 612 steps and 3 repeats.

Two torsos conjoinment

Now on Anima. The poses are too rigid, but changing the weight on 2girls and two torsos to 0.6 and the lora strength to 0.8 solves this.

Four_heads_conjoinment_v1-an_anima_3397602_epoch_8.safetensors

Two_torsos_conjoinment_v1-an_anima_3395539_epoch_8.safetensors

v1.0-anima

v1 front-back

This does not feel very necessary, but i feel like i need it for some things, so i made it. This lora is also for 5 and 6 heads, but may not be as easy to generate.This lora is a bit quirky. 4 heads often makes 3, 5 heads often makes four or six, and 6 heads often makes 4 or 5.4 heads needs four heads, 4girls, (2heads:1.3), broad shoulders.5 heads needs five heads, 5girls, (3heads:1.2), broad shoulders or (six heads:1.3), (6+girls:1.3), 2heads, broad shoulders. But that makes 4 or 6 sometimes.6 heads needs (six heads:1.4), (6+girls:1.4), 2heads, conjoined, broad shoulders, but it makes 5 often.conjoined is not always necessary but does have an effect.Use a 1.2 strength.

Four heads conjoinment

I'm suprised with how well this lora handles six heads.This lora often puts too many heads or too few. For four heads you need to reduce both the girls and the heads triggers to 0.6 to solve it. Five heads needs both the girls and the heads triggers on 1.3, broad shoulders tag, and wide resolution. Six heads pics need the six heads and the 6+girls tags on normal weight, broad shoulders tag, and wide resolution. Body is too wide i guess.The "broad shoulders" tag helps a lot, but is not needed in wide resolutions.

Files with SHA256: 2987368ab7a61c6c7d71cf29bb7c249fe2d5554c8ee26f1802c36d215f0d0683