<h2 id="anima-v2-mix">Anima V2 Mix</h2>The origional LoRA tended towards anime when using boruu tags alone. So I tried to retrain with a cleaned dataset and wd14 tagging - the result was a bit to intense. So i tried to make e new and much shorter natural language captioning - with a mixed result for me. So I took bothe and fiddled araound with a LoRA merger and mixed them. This is why both Triggers work:@biomechstyle for the nat language@biomechanical for the wd14 oneHave fun and post some Images :-)Maybe I'll release the other two variants to play around with.________________________________________________________________________________________________Trained with 100 hand picked synthetic images, mostly indoor biomechanical environments (no humans, no sexual content). The captioning was made with Qwen 3VL in natural language - so you must be descriptive if aiming for a similar style - pure booru tags do not work as good. Just have a look at the example pictures. Besides not knowing how a biomechanical humans should look like - it can come close to common biomechanical art styles and games and I would suggest it is more suitable for mature themes. For now I can say that the folowing tags have a strong influence: biomechanical style,depth of field,grotesque,eerie, unsettling,blending X with Y, (for X and Y use something like "biologic" "industial" "organic" "metallic")0.8 is a good starting point

Painterly Biomechanical

now for MOSS - trained on base model with ai-toolkit.Sorry for the missing Prompts - just DL the image and put it into Comfy. I used the most basic workflow - I*ll upload new ones later.

Files with SHA256: 7d0ec7ef8cfc00616da4759fc5827bb7f94201e023a3af96c481f2cf8dddb8b0